Home
"An Overview of Gene Identification: Approaches
Contents
1. actual predicted Predicted Positives PP Negatives PN Positives AP TP FN Actual Negatives AN FP TN Sensitivity Sn TP TP FN Specificity Sp TP TP FP TP TN FP FN Correlation coefficient CC PP PN AP AN Figure 4 1 2 Sensitivity vs specificity In the upper portion of the figure the four possible outcomes of a prediction are shown a true positive TP a true negative TN a false positive FP and a false negative FN The matrix at the bottom of the figure shows how both sensitivity and specificity are determined from these four possible outcomes giving a tangible measure of the effectiveness of any gene prediction method Figure adapted from Burset and Guig 1996 and Snyder and Stormo 1997 For any given prediction there are four possible outcomes detection of a true positive a true negative a false positive or a false negative Fig 4 1 2 Two measures of accuracy can be calculated based on the ratios of these occurrences a sensitivity value reflecting the fraction of actual coding regions that are correctly predicted as being coding regions and a specificity value reflecting the overall fraction of the prediction that is correct In the best case scenario the methods will optimize the balance between sensitivity and specificity in order to be able to find all true exons without becoming so sensitive as to pick up an inordinate number of
2. May Feedback Life Sciences Divisian ORNL Disclaimer Wobmacter ET menet E Figure 4 9 2 Pipeline submission form different formats When an analysis module fails instead of the green Succeeded icon the user would see a red Failed icon for that module For subsequent analysis modules that are dependent on the failed analysis module the user would see a red Aborted icon Subsequent analysis modules that are independent of the failed module would be run normally The Summary page also provides three useful links for retrieval and display of the pipeline results The Raw output link allows the user to retrieve the pipeline results as a single text document Individual analysis output can be viewed in Raw or Text Table format by selecting the appropriate menu option from the pull down menu provided next to the Status icon for that particular analysis refer to Fig 4 9 3 Figure 4 9 4 illustrates the Text Table output for GrailEXP Gene Finder analysis It consists of a summarized list of genes followed by detailed information about each predicted gene and its components consist ing of exons promoters and poly A sites This is followed by the list of protein and mRNA sequences for each of the predicted GrailEXP genes Current Protocols in Bioinformatics Finding Genes 4 9 5 Supplement 4 GrailEXP and Genome Analysis Pipeline for Genome Annotation 4 9 6 Supplement 4 Tools
3. Mp opm y JA A N 69 0o Eukaryotic GeneMark hmMmm masss om pane Reference Borodovsky M and Lukashin A unpublished 10 2002 O sativa Rice Eukaryotic GeneMark hmm model has been updated pdotes Input Sequence Tithe optional Sequence e TTT GAT TTT SPOS Tete TOTGCTTTATTAGTTAATTASTAOCCATT ICT OAAGAAGAAAT AACA TAA Sequence File upload e Browse Species o Maapiers Output Options Email Address required for graphical output or sequences longer than 100000 bp e neveneridenber brokazy garehe Generate PostScript graphics Print GeneMark 2 4 predictions in addition to GeneMark hmm predictions Translate predicted genes into protein e Run Ootaut Start Gone Mark rm Figure 4 6 1 The user interface for the eukaryotic GeneMark hmm program Required input includes a DNA sequence in FASTA format either copied and pasted into the Sequence text box or uploaded from one s local drive using the Browse button next to the Sequence File Upload box downloaded as file M12532 fna at http www3 interscience wiley com c_p cpbi_sampledatafiles htm The sequence can be pasted into a sequence window if shorter than 100 kb or uploaded as a file if longer The last exon of this gene shown in bold in Figure 4 6 3 is alternatively spliced as either human serum albumin or human alloalbumin Venezia 1 Via a Web browser connect to http opal biology gatech edu GeneMark eukhmm cgi In the Input Sequ
4. and modify the value Note that the memory requirements of the program go up as this value is increased Another way is to break down each long gt 4 Mb sequence into shorter ones RepeatMasker does not fail explicitly even if one s hard disk is full it actually gives ap parently normal results Therefore when it is noticed that the results are far from those ex pected there might be a disk space problem Using q or qq see Basic Protocol 2 step 8 can speed things up but the sensitivity is reduced When WU BLAST is used the s slow option is preferred since the speed with WU BLAST is reasonably fast and the mask ing results are better Analysis on smaller sequences lt 2 kb could be less accurate Note that previous version s of Repeat Masker had a problem with overwriting the files with the same names when multiple anal yses were performed on the same input files This is no longer a problem since Repeat Masker creates output directories for each analysis Literature Cited Bao Z and Eddy S R 2002 Automated de novo identification of repeat sequence families in sequenced genomes Genome Res 12 1269 1276 Bedell J A Korf I and Gish W 2000 MaskerAid A performance enhancement to RepeatMasker Bioinformatics 16 1040 1041 Jurka J 2001 Repbase update a database and an electronic journal of repetitive elements Trends Genet 16 418 420 Jurka J Kapitonov V V Pavlicek A Klonowsk
5. After this point the sequences are too differ ent to align well over large stretches of the genome and the percentage of mismatches and gaps decreases R Brown and M Brent unpub observ Errors in the target sequence or the assem bly can have a major impact on the perfor mance of gene predictions Particularly bad errors include frame shifts and the introduc tion or alteration of stop codons or splice sites For TWINSCAN and N SCAN the quality and continuity of the informant database is less important the accuracy of gene predic tion differs only marginally when a database of shotgun reads with 3 x coverage is used in stead of 7x coverage assembled into contigu ous sequences R Brown unpub observ Overall recommendations The best gene prediction method to use de pends a great deal on the application One of the main considerations should be the value of increasing sensitivity as compared to the cost of decreasing specificity Sensitivity can often be increased by including predictions by any one of multiple systems whereas speci ficity can be increased by considering only predictions made by all systems Guigo et al 2003 For high value projects focusing on a few kilobases of sequence multiple meth ods can be used and the results can be com bined either by manual inspection or by spe cific combiner programs such as JIGSAW Allen and Salzberg 2005 and GLEAN Elsik et al 2007 However for high
6. Copy and paste gene prediction Submissions by guest Submission Submit Date Sequence Genome Submission 63 2007 03 26 chr2 human Submission 62 2007 03 23 chr2 human ubmission 61 2007 03 23 chr2 human Submission 60 2007 03 22 chr2L D melanogaster Submission 56 2007 03 22 chr2L D melanogaster Submission 55 2007 03 22 chr2L D melanogaster Submission 53 2007 03 21 ch2L D melanogaster Submission 52 2007 03 21 chr2L D melanogaster Submission 51 2007 03 20 chr2L D melanogaster Submission 50 2007 03 20 chr2L D melanogaster Submission 49 2007 03 20 chr2L D melanogaster Submission 48 2007 03 19 chr2 human Total Submissions 12 Copyright 2007 Questions Comments help mblab wusti edu Figure 4 8 5 An example of a My Submissions page From here all previous and running jobs can be accessed using the links on the left Finding Genes 4 8 7 Current Protocols in Bioinformatics Supplement 20 ALTERNATE PROTOCOL 1 Using N SCAN or TWINSCAN 4 8 8 Supplement 20 RUN N SCAN FROM THE COMMAND LINE ON A LOCAL COMPUTER Running N SCAN on a local computer as opposed to running it via the Web server see Basic Protocol is recommended for users with bioinformatics experience who need to 1 process gt 1 Mb per day on a sustained basis 2 use N SCAN outside the supported clades or 3 use N SCAN on proprietary sequences The N SCAN software package includes the N SCAN executable itself together with
7. Elerent _ Element Fra Len Ser Type Begin End promoter 1978 2065 70 0 69 exon 3806 4063 i 168 1 00 pen 29990 20901 2 52 Stith Figure 4 9 4 GrailEXP genes text table The Java Viewer applet uses Sun s Java 2 plugin If the user does not have the plugin installed upon loading the Java Viewer page the browser will automatically point to Sun s Java Plugin download page and prompt the user for permission to download and install the plugin If the user agrees automatic plugin download will be initiated If there is a problem with automatic installation the user can manually download the plugin from Sun s Web site at http java sun com getjava download html and install it The Java Viewer is under active development at the time of this writing New functionality is being added and therefore the user may have access to more features than are covered above Since Java itself is undergoing rapid development the Java Viewer may be modified to take advantage of newer features Therefore it may be necessary to install a new version of the plugin The pipeline Web page will be updated to provide information about the current and expected enhancements as well as User s Guide and Help sections Current Protocols in Bioinformatics Finding Genes 4 9 7 Supplement 4 fe JPipeDisplay for RequestiD 1059421518 28565 Jet Java Applet Window Ii Sts 1 root 14001 71001 20001 35001 Grad XP Genas Gens
8. The sample sequence example fna which contains region 1 to 50 000 from Escherichia coli K12 used to illustrate this protocol can be downloaded from the Current Protocols Web site http www3 interscience wiley com c_p cpbi_sample datafiles htm 1 Via a Web browser connect to http opal biology gatech edu GeneMark gmhmm2_ prok cgi In the Input Sequence section paste an input sequence into the Sequence box area or alternatively click on Browse next to the Sequence File Upload box to upload the input sequence file from a local drive Current Protocols in Bioinformatics Finding Genes 4 5 7 Supplement 1 J Genclik hmm for Prokaryotes Microsofti Internet xplorer tie Gd ye Pevetes Jock p Q Q a pen toos Grete C tes Kona betay geh Af arerahipt mm prh lt a GeneMark hmm for Prokaryotes and Low Eukaryotes Version 2 1 orm pase Reference Lukashin A and Borodovsky M GeneMark hmm new solutions for gene finding NAR 1998 Vol 26 No 4 pp 1107 1115 Download POF y This page has been updated to run version 2 1 of Ganemark hmm for prokaryotes ss well as version 2 4 of GeneMark This version is significantly faster that the previous version and can analyze amp sequence of 1 million base pairs in under 15 seconds Support of prediction of overlapping genes has been vastly improved as weill 0 Please note thet email is the only way to receive results for sequences longer than 4 MB oe This
9. dl ee E com uan to i PA EE A E EAT GS PERN EEN G E OEN GP EI OOE PER BEN D TERE RORY SEE NESEY EA S 8800 9200 9600 10000 10400 Nucleotide Position gm_sequence_typical_ps om_sequence_atypoal p Figure 4 5 6 The graphical output from the GeneMark hmm program for the sample sequence The format is the same as the GeneMark graphical output with the genes predicted by the Typical model in black solid line and the genes predicted by the Atypical model in red dashed line The wide black horizontal lines indicate regions predicted by GeneMark hmm as protein coding 3 Scroll further down the page and set the Output Options The user may request the graphical output Also GeneMark predictions can be requested in addition to those of GeneMark hmm An E mail address is required for sending text output for sequences longer than 4 000 000 nt or if graphical output is requested 4 After completing the above entries click the Start GeneMark hmm button The results will be depicted on the browser or will be sent to the E mail address provided Finding Genes 4 5 9 Current Protocols in Bioinformatics Supplement 1 BASIC PROTOCOL 2 Prokaryotic Gene Prediction Using GeneMark and GeneMark hmm 4 5 10 Supplement 1 5 Interpret the text output The GeneMark hmm text output Fig 4 5 5 contains a listing of all regions predicted as protein coding Genes predicted on the direct strand are indicated with
10. ksh ksh Figure 4 6 7 Specifying the matrix file name and path in the environmental variables DEFMAT_HMME and MATPATH in Unix and in ksh shells GeneMark hmm Version 2 2 Sequence length 19002 bp G C content 35 02 Thu Jul 25 16 04 48 2002 Predicted genes exons Gene Exon Strand Exon Type Internal Internal Internal Internal Internal Internal Internal Internal Internal Internal Internal Internal Internal WO nnhbwne 10 1l 12 13 Pe ee e e e ee ee D ee Predicted gene sequence s a Sequence name HALB gb fna Exon Range 1617 2564 4076 6041 6802 7759 9444 10867 12481 13702 14977 15534 16941 d 1854 2621 4208 6252 6934 7856 9573 11081 12613 13799 15115 15757 17073 gt HALB gb fnalGeneMark hmm gene 1 581 aa SAYS RGVFRRDAHKS EVAHRFKDLGEENF KALVLIAFAQYLQQCPFEDHVKLVNEVTEFA KTCVAD ESAENCD KSLHTLFGDKLCTVATLRET YGEMA DC CAKQE PE RNECFLQHKDD NP NLPRLVRPEVDVMCTAFHDNEETFLKKYL YE IARRHPYFYAPELLFFAKRYKAAFTECCOQ AADKAACLL PKLDEL RDEGKASSAKQRLKCASL OKF GE RAFKAWAVARLSOQRFPKAEFPAE VSKLVTDLTKVHTECCHGDLLECADDRADLAKY ICENQDS ISSKLKECCEKPLLEKSHCI AEVE NDEMPADLPSLAADFVESKDVC KNYAEAKDVFLGMFLYE YARRHPDYSVVLLLRLA KTYE TTLEKCCAAAD PHECYAKVF DEFKPLVEE PONLIKQNCELFEQLGEYKFOQNALLVR YTKKVP OVS TP TLVEVSRNLGKVGSKCCKHP EAKRMPCAEDYLSVVLNOLCVLHEKTPVS DRVTKCCTESLVNRRPCFSALEVDETYVP KE FNAETFTFHADICTLS EKERQIKKQTALYV ELVKHKPKATKEQLKAVMDD PAAF VE KCC KADDKETCFAEE Matrices file MTX
11. pull down menu located below the Gawain Gene Models check box allows the user to limit the use of alignments to sequences of only the specified organism Also see Background Information for discussion of Gawain 7 Choose whether or not to locate CpG islands Checking the box labeled CpG Islands will locate CpG Islands within the DNA sequence using the Grail 1 3 CpG island location algorithm 8 Choose whether or not to locate repetitive elements and mask the sequence for repetitives Checking the box labeled Repetitive Elements will locate simple repeats using the GRAIL 1 3 simple repeat location algorithm It will also locate complex repeats by performing a BLAST search against the RepeatMasker A FA Smit and P Green unpub observ http ftp genome washington edu RM RepeatMasker html database The sequence corre sponding to the located complex repeats will be masked and returned in the browser window Input sequence and run analysis 9 Input the DNA sequence Cut and paste the DNA sequence in the window labeled DNA Sequence or click on the Browse button to select a file on the local computer drives for uploading A valid FASTA format or Raw text sequence must be provided to the program The sequence must be at least 100 bases in length and no more than 500 kb It should be emphasized that GenBank or other formats will not work correctly The sequence may either be cut and pasted into the window or uploaded from a file Clicking
12. 2 jah fevertes ieda 7 Nite Nene Detgy gated Apana gremat lt op GeneMark Version 2 4 isisi ts noon Reference Borodoveky M and McIninch J GeneMark parallel gene recognition for both DNA strands Computers amp Chemistry 1993 Vol 17 No 19 pp 123 133 Web site hes been redesigned and moved a to new more powerful server ihe Of previous veda bes Input Sequence Title optianal Sequence Sequence File uplosd e Running Options Species o E co Window size e bp RBS model e Nore Step sizece 12 bp Use alternate genetic code e Threshold e 05 C Eukeryote 6 9 Yeast ATG only start Mycoplesme TGA Tryptophan Output Options Graphical output oprionse Text output optionse O Generate PostScript graphics A List open reading frames ORFs predicted as coding Mark orfs on graph sequences CDSs B Mark regions on graph amp List regions of interest O Mark stop codons on graph List putative eukaryotic splice sites O Mark start codons on graph D Write protein translations of ORFs O Mark frameshifts on graph Write nucleotide transcripts of ORFs Mark putative exon splice sites Write protein translations of regions O Print graph in landscape format O Write nucleotide transcripts of regions O Write protein translations of putative exons Email address required for graphical Verite nucleotide transcripts of putative exons output Run Ceiau Stat Genedar Q tere Figure 4 5
13. 392 18 lt 02 275 63 90 s19 48 exon Figure 4 2 4 Prediction results from AAT Note that all the predictions are on the forward strand and that no exons are predicted on the reverse strand PO or Prior probability It reflects the a priori belief on the coding exon density in the genomic region As one can see from the above examples when PO was changed from 0 02 to 0 04 MZEF predicted two more exons that include one true exon 2564 2621 and another false positive exon 17812 17874 see Figures 4 2 1 and 4 2 3 So the effect of increasing PO is to have more putative exons predicted The default value is 0 02 for the Web version and is 0 04 for the local version Overlap allows predicted exons to overlap The default is 0 namely overlapping is not allowed As shown in the command line version example above when Overlap 1 is set at most one overlapping exon is allowed to output for each exon region see Figure 4 2 2 This allows the user to choose an exon with an alternative splice site especially when one is looking for an exon that has a compatible frame with other adjacent exons during gene model building Normally if G C_content is low the exon density may also be low Current Protocols in Bioinformatics Finding Genes 4 2 11 Using MZEF to Find Internal Coding Exons 4 2 12 Results The predicted exon boundaries by MZEF are further characterized by Splice Pr
14. 656 111 10 561 0 557 0 689 4076 4208 0 985 0 405 0 322 0 567 221 10 5360 5050 587 6041 6252 0 94210 391 0 596 0 513 212 0 5380 573 0 646 6802 6934 0 8690 363 0 553 0 522 211 10 5470 51810 545 7759 7856 0 99910 64210 59210 481 112 0 5360 622 0 607 9444 9573 10 99810 636 0 486 0 463 122 0 574 0 581 0 553 10867 11081 0 994 0 626 0 493 0 403 122 0 541 0 558 0 597 12481 1261310 998 0 453 0 617 0 544 212 0 576 0 588 0 548 13341 13425 0 604 0 41410 4980 556 221 0 460 0 537 0 719 113702 1379911 000 10 587 l0 482 0 399 122 0 5480 532 0 719 114977 151150 757 0 502 0 591 0 509 212 0 526 0 574 0 457 115534 15757 0 508 0 501 0 463 0 594 221 0 462 0 570 0 562 116941 17073 0 998 0 667 0 451 0 515 122 0 48310 613 0 609 IE Done IO itera r Figure 4 2 1 The screen dump from an example run using M12523 fasta as the input sequence with all the default parameters CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 17688 17732 that may be compared with the MZEF predictions below 1 Using a Web browser connect to http Avww cshl org genefinder select the Human button and cut and paste the FASTA sequence maximum 200 kb into the input window Alternatively type in th
15. Adams M D Clayton R A Fleischmann R D Bult C J Kerlavage A R Sutton G Kelley J M Fritchman J L Weidman J F Small K V San dusky M Fuhrmann J L Nguyen D T Utter Finding Genes 4 5 15 Supplement 1 Prokaryotic Gene Prediction Using GeneMark and GeneMark hmm 4 5 16 Supplement 1 back T R Saudek D M Phillips C A Mer rick J M Tomb J F Dougherty B A Bott K F Hu P C Lucier T S Peterson S N Smith H O Hutchison M C A and Venter J C 1995 The minimal gene complement of Mycoplasma genitalium Science 2170 397 403 Hayes W and Borodovsky M 1998a How to in terpret anonymous genome Machine learning approach to gene identification Genome Res 8 1154 1171 Hayes W S and Borodovsky M 1998b Deriving Ribosome Binding Site RBS statistical models from unannotated DNA sequences and the use of the RBS model for N terminal prediction In Proceedings of Pacific Symposium on Biocom puting 1998 pp 279 290 World Scientific Press Singapore Lawrence C E Altschul S F Boguski M S Liu J S Neuwald A F and Wootton J C 1993 De tecting subtle sequence signals A Gibbs sam pling strategy for multiple alignment Science 262 208 214 Lukashin A V and Borodovsky M 1998 Gene Mark hmm New solutions for gene finding Nu cleic Acids Res 26 1107 1115 Tatusov R L Mushegian A R Bork P Brown N P Hayes W Borodo
16. Finding Genes 4 5 3 Current Protocols in Bioinformatics Supplement 1 DD WebGenetark Microrett internet Explorer Fie Gm Yew Ppots pds b Om gt a A PAE i Fates a C 2 ta Mops Deotegy gate eterna Mant igermmant tcp GeneMark Version 2 4 soss iis page Reference Borodovsky M and Mcininch 3 GeneMark parallel gene recognition for both ONA strands Computers amp Chemistry 1993 Vol 17 No 19 pp 123 133 i Web site has been redesigned and moved a to new more powerful server rie ot preno updates Result of last submittal GeneMark Results eo to R edicted OFF s Go to List of predicted regions of intereat Sequence gt Escherichia coli Kil Pegien i 50000 Sequence length 0000 OC Comment 32 194 Vindow length P Vandow step it Taresheid value 0 500 Betcix E cols NCBI FP 3 Order 4 Matsix author JON Amiga TreansNatesx atrix order 4 List of Open reading frames predicted as CDSs shown with alternate starts regions fros stert to stop codon v coding function gt 0 50 Lett Fight ota Coding Ave Start Frase Prob Prob direct tr direct tr direct tr ditect fr direct fr gt 0 39 gt on 4 0 85 4 0 05 gt o 9 6 Ld eee gt gt gt gt direct tr direct te direct tr aireet tr direct fr ETTEI dicect tr direct tr direct tr dicect fr to te te te complement fr complement fr complement fr complement fr
17. J D Kerlavage A R Dougherty B A Tomb J F Adams M D Reich C I Overbeek R Kirk ness E F Weinstock K G Merrick J M Glodek A Scott J L Geoghagen N S M Weidman J F Fuhrmann J L Nguyen D Ut terback T R Kelley J M Peterson J D Sa dow P W Hanna M C Cotton M D Roberts K M Hurst M A Kaine B P Borodovsky M Klenk H P Fraser C M Smith H O Woese C R Venter J C 1996 Complete genome se quence of the methanogenic archaeon Methano coccus jannaschii Science 273 1058 1073 Durbin R Eddy S Krough A and Mitchison G 1998 Biological sequence analysis Prob abilistic models of proteins and nucleic acids Cambridge University Press Cambridge Fleischmann R D Adams M D White O Clay ton R A Kirkness E F Kerlavage A R Bult C J Tomb J F Dougherty B A Merrick J M McKenney K Sutton G Fitzhugh W Fields C A Gocayne J D Scott J D Shirley R Liu L I Glodek A Kelley J M Weidman J F Phillips C A Spriggs T Hedblom E Cotton M D Utterback T R Hanna M C Nguyen D T Saudek D M Brandon R C Fine L D Fritchman J L Fuhrmann J L Geoghagen N S M Gnehm C L McDonald L A Small K V Fraser C M Smith H O and Venter J C 1995 Whole genome random se quencing and assembly of Haemophilus influen zae Rd Science 269 496 5 12 Fraser C M Gocayne J D White O
18. The BLAST searches UNITS 3 3 amp 3 4 are less likely to hang when searching smaller data bases The Galahad alignment phase is run in parallel In either mode BLAST can of course run in its multithreaded mode The parallel search is accomplished through the use of TCP IP client server system using Daniel Bernstein s ucspi tcp program thus avoiding the overhead of PVM or MPI implementation The use of BLAST for obtaining initial ap proximate alignments rather than the Smith Waterman like algorithm used by some other systems provides immense speedup GrailEXP refines these alignments in a subsequent phase A drawback of using BLAST is that it does not look for short alignments This means that find ing short internal exons must be done by the Galahad alignment program which is currently being refined for this purpose The BLAST simple repeat filtering option causes a break in otherwise good alignments Ideally BLAST should filter for simple repeats initially but then de filter to join fragments together that it incorrectly split Galahad attempts to join to gether such breaks provided the length of the break is less than 200 bases Galahad uses dynamic programming to pick the best alignments such that those with the highest percent identity get joined together This causes incorrect alignments to be reported in the case of multi exon repeating genes such as back to back to back zinc fingers in close proximi
19. UNIT4 2 GENSCAN XGRAIL unir4 9 and PowerBLAST Zhang and Madden 1997 was used in an integrated fashion in the prediction of gene structure Kuehl et al 1999 Another integrated approach to this approach involves workbenches such as Genotator that allow users to run a number of prediction methods and homology searches simulta neously as well as to annotate sequence features through a graphical user interface Harris 1997 A combinatorial method developed at the National Human Genome Research Institute links most of the methods described in this chapter into a single tool This tool named GeneMachine allows users to query multiple exon and gene prediction programs in an automated fashion Makalowska et al 1999 A suite of Perl modules are used to run MZEF GENSCAN GRAIL2 FGENES and BLAST RepeatMasker UNIT 4 10 and Sputnik are used to find repeats within the query sequence Once GeneMachine is run a file is written that can subsequently be opened using NCBI Sequin in essence using Current Protocols in Bioinformatics File Edit Search Options Misc Annotate Target Sequence bigcont Done play Forsa Graphic ty GeneMachine a 3o 20503 4 vimota t A 4 A A102620 aA 2 A eat 15154 x x x gt 6 mio w o me b gt 4 at 2 o 4 ue gt Diest wiscprot t amp TET TELA u PELI Fgenes F genes gt Grol good Genscan 5 gt v evon MZEF P0457 exom MZF P0425 exon Gra
20. file 7 When filtering is used the false positive and false negative rates are given only for a default length of the filter window 60 bp thus any change in the length of either of the filter windows i e parameter on line 10 or 11 of the config file Table Current Protocols in Bioinformatics Finding Genes 4 4 9 A 11 090 j predator apertea Malaria train train11_01 TrainG imat2001 11 09014 34 57 nore false nofilter acc Thr False negatives positives 15 00 05 11 9 OONA ANNEO 5 5 7 9 3 9 0 Figure 4 4 4 Example of false nofilter acc file z nm2o01 11 090 predator apertea Halaria train train11_01 TrainG imal2001 11 09914 34 57 more false nofilter don False 0 ia 222 ER gt e e or oJ oO WO REVS HSREH ae 927 Oc WOON MOOS Wh e PPR oe amp z y fe y amp y fo y fe y A y amp y y fe y ra fe y fe y amp y amp ee ee ee eee eee eee eee ee a 4 801180 More 82 at Figure 4 4 5 Example of false nofilter don file 4 4 10 Current Protocols in Bioinformatics 4 4 2 will cause a change in the value of the corresponding threshold i e parameter on line 12 or 13 of the config file Table 4 4 2 Therefore re run the train GlimmerM procedure using the a and d optional parameters after changing the length of the filter window for either donor or acceptor sites s
21. influencing chromosomal structure and function will obviously be of great importance in applying human sequence data to better understanding human biology and possibly predicting potential disease risks NHGRI has launched a new consortium based effort to determine the parts list through an effort called ENCODE the Encyclopedia of DNA Elements The effort will concentrate on a representative 1 of the human genome and is intended to identify all of the functional elements in the human genome sequence identify gaps in our ability to annotate the sequence and consider the applicability of newly developed methods in analyzing the entire genome One of the methods discussed in this chapter provides an excellent example of the kinds of approaches that can be used to annotate whole genomes TWINSCAN uni7 4 8 relies on target genome parameters information from the sequence to be annotated con servation parameters information from closely related sequences and BLAST pa rameters to predict genes within the genomic region of interest The approach has already been used to annotate the human and mouse genomes and the results of these predictions can be found through the UCSC Genome Browser UNIT 1 4 Approaches such as this as well as those described throughout this chapter will be critical not only in identifying the parts list but in fulfilling the promise of using genomic information in a way that can guide both basic a
22. naw solutions for gene finding NAR 1998 Vol 26 No 4 pp 1107 1115 Download POF t This page hes been updated to run version 2 1 of Genemark hmm for prokaryotes as well as version 2 4 of GeneMark This version is significantly faster that the previous version and can analyze a sequence of 1 million base pairs in under 15 seconds Support of prediction of overlapping genes has been vastly improved as well 0 Please note that email is the only way to receive results for sequences longer than 4 MB This service is in a testing phase Please report problems and offer suggestions to John Besemer New models included for many newly sequenced prokaryotic genomes ng of orenous update Input Sequence Tithe optional e Sequence Texte Sequence File upload e Species Escherichia ook KI Typical Model BAtypical Model Output Options E Mail Address required for graphical output or sequences longer than 4000000 bp e Generate PostScript graphics Print GeneMark 2 4 predictions in addition to GeneMark henen predictions o Run a Figure 4 5 4 The user interface for the GeneMark hmm program Required input includes a DNA sequence either copied and pasted into the text box or uploaded as a file from the user s computer in FASTA format the selection of the correct species model and the selection of the Typical model the Atypical model or both Files A single sequence in FASTA format APPENDIX 1B
23. or more related informant genome s Such a multi genome de novo system is generally better than a similar system run with only one genome For any given genome how ever a well optimized single genome predic tor can equal or surpass a poorly optimized dual genome predictor Annotation pipeline methods for gene pre diction use databases of known proteins and or cDNA sequences and map these to the tar get genome in an attempt to predict genes whose protein products would be similar to those of known genes Subsequently de novo annotation is attempted for regions that have no matching sequences The best known expression based system is Ensembl Hubbard et al 2007 Such systems have the advan tage of providing a single discrete piece of evidence for each prediction namely similar ity to a specific cDNA or its translation and hence are sometimes called evidence based Mouse Genome Sequencing Consortium et al 2002 However the mapping problem is difficult so accuracy is not perfect even for close relatives of known genes Furthermore these pipelines still need a reliable de novo predictor for unmatched regions or leave them unannotated Accuracy measures The accuracy of gene predictors can be as sessed by their ability to predict individual coding nucleotides exons either exactly or approximately and complete genes and also by a variety of other measures Guigo et al 2000 For all these measures it
24. overlap 1 A copy of the FASTA file for the DNA sequence of interest e g m12523 fasta see Necessary Resources must be copied into the MZEF directory For information on navi gating through a Unix environment see APPENDIX IC If you intend to download the program and its associated data files for more than one organism the directories should be named in a way that the user can keep track of the files e g MZEF_HUMAN in the case of the human data set 2 Download and install the appropriate MZEF executable file and all of the required data files All of the files are available by running an FTP session as follows sftp cshl org anonymous Name Password ftp gt ftp gt ftp gt ftp gt ftp gt ftp gt answer yes to all the files this will download the required data files ftp gt The instructions on how to install MZEF are in the README file which also has a brief description of the program and parameters The command get mzef_cmd_1mb_sun mzef_cmd downloads the executable your internet address cd pub science mzhanglab mzef get README cd human binary get mzef cmd_ilmb_sun mzef_cmd mget quit mzef_cmd_1mb_sun and renames it mzef_ cmd Current Protocols in Bioinformatics Finding Genes 4 2 5 ALTERNATE PROTOCOL Using MZEF to Find Internal Coding Exons 4 2 6 3 Change the permissions on the executable by issuing the following command chmod r
25. throughput projects involving megabases of sequence this may not be feasible Another consideration is the importance of finding genes that are truly novel If finding novel genes without sacrificing specificity is impor tant N SCAN is the method of choice for most of the genomes described above If find ing novel genes is important but sensitivity is more important than specificity other gene finders may be better for mammals Finally if accurate annotation of genes that are similar to known genes is most important a method like Ensembl may be best N SCAN versions and new features N SCAN is under active development A Web site for TWINSCAN the previous im plementation was created in 2003 N SCAN was published in 2006 and contains consid erable improvements N SCAN has a more sophisticated evolutionary model than TWIN SCAN in that it considers the specific bases in the alignment rather than treating all mis matches and deletions as identical It can also use alignments of more than two sequences and model their phylogenetic relationships N SCAN predicts 5 UTRs including com pletely non coding exons and it can use align ments of expressed sequence tags ESTs to improve accuracy N SCAN can also be run in TWINSCAN mode TWINSCAN mode uses a different representation of alignments called conservation sequence which is com posed of three symbols denoting match mis match deletion or unaligned region and dif feren
26. 1 The user interface for the GeneMark program Required input includes a DNA sequence either copied and pasted into the text box or uploaded as a file from the user s computer in FASTA format and the selection of the correct species model Other options are available according to the interest of the individual user Necessary Resources Hardware A personal computer or workstation with Web access Software A Web browser Files A single sequence in FASTA format APPENDIX 1B The sample sequence example fna which contains region 1 to 50 000 from Escherichia coli Prokaryotic Gene K12 used to illustrate this protocol can be downloaded from the Current paneer ene Protocols Web site http www3 interscience wiley com c_p cpbi_sample GeneMark hmm datafiles htm 4 5 2 Supplement 1 Current Protocols in Bioinformatics 1 Via a Web browser connect to http opal biology gatech edu GeneMark genemark24 cgi In the Input Sequence section paste an input sequence into the Sequence box area or alternatively click on Browse next to the Sequence File Upload box to upload the input sequence file from a local drive The Sequence File Upload option is more powerful since the copy and paste method imposes a limit on the length of the sequence If the sequence has a FASTA APPENDIX 1B title line e g gt Sequence name this name will be assigned to the sequence in the output unless the user gives a name in the Sequence Title text a
27. 6 hgi d 6635 6763 15776 LIPAN LME L 6034 6163 5 11 sos 9 3 0 8 0 0 hgi8_dna 6884 7043 15496 MIR SINE MIR 79 250 qs 12 l 400 30 5 9 4 1 7 hgi _dna 7064 7184 15355 MIRb SINE MIR 140 262 60 213 327 32 5 2 5 0 8 hhgi8_dna 7260 7500 15039 C MIRe SINE MIR 8 260 19 14 ls 383 34 2 4 6 4 1 bgi8_dna 9370 9504 13035 HIR SINEAD 90 226 36 15 2 282 22 8 7 4 5 80 hgi8_dna 9612 9730 12809 C MIR SINE MIR to 262 124 16 l 270 31 1 16 7 0 7 bgo 9798 9995 12544 MIR SINE NIR 21 2 404 32 4 7 2 5 0 hgi _dna 10016 10067 12472 GA rich Low_complexity 1 52 0 18 hgia_ 0123 10261 12278 C MIR s MIR 47 215 63 19 373 27 7 42 f hgi _ dna 10641 10780 11759 MIRc SINE MIR 101 238 24 20 hgi _ dna 12043 12314 10225 C MERI21 DRA TcMar 37 360 76 21 572 29 6 7 3 2 5 hgi _ dna 13353 13529 9010 C MIRb SINE MIR 58 210 26 22 ls 300 32 2 6 2 1 6 hgi _dna 13549 14201 8338 LIMESA LINB LI S461 6127 46 23 2277 26 6 3 2 11 harif Ana 14948 14449 SH77T1 C LIMI LIME Lt ann ania anar 24 ls 1676 14 6 1 8 1 7 Figure 4 10 2 Web RepeatMasker result from an example run showing the repetitive elements annotations section which lists cross_match summary lines this result is available in Text File Format A and XHTML format B See Guidelines for Understanding Results and Table 4 10 1 for explanation Finding Genes 4 10 3 Current Protocols in Bioinformatics Supplemen
28. A workbench for sequence annotation Genome Res 7 754 762 International Human Genome Sequencing Consortium 2001 Initial sequencing and analysis of the human genome Nature 409 860 921 Kuehl P Weisemann J Touchman J Green E and Boguski M 1999 An effective approach for analyzing prefinished genomic sequence data Genome Res 9 189 294 Liu A Y Torchia B S Migeon B R and Siliciano R F 1997 The human NTT gene Identification of a novel 17 kb noncoding nuclear RNA expressed in activated CD4 T cells Genomics 39 171 284 Current Protocols in Bioinformatics Makalowska I Ryan J and Baxevanis A 1999 GeneMachine A unified solution for performing content based site based and comparative gene prediction methods 12th Cold Spring Harbor Meeting on genome mapping sequencing and Biology Cold Spring Harbor NY Makalowska I Sood R Faruque M U Hu P Eddings E M Mestre J D Baxevanis A D and Carpten J D 2002 Identification of six novel genes by experimental validation of GeneMachine predicted genes Gene 284 203 213 Pearson W R Wood T Zhang Z and Miller W 1997 Comparison of DNA sequences with protein sequences Genomics 46 24 36 Rogic S Mackworth A and Ouellette B F F 2001 Evaluation of Gene Finding Programs Genome Res 11 817 832 Snyder E E and Stormo G D 1993 Identification of coding regions in genomic DNA sequences An application of dynamic programmi
29. BLASTP Altschul et al 1990 unrr 3 3 to make func tional prediction for the putative genes Experi mental biologists can use the sequences around the predicted exons and genes to create primers for PCR analysis of genes of interest as well as to design oligonucleotides representing pro tein coding regions for DNA expression arrays Literature Cited Altschul S F Gish W et al 1990 Basic local alignment search tool J Mol Biol 215 403 410 Besemer J Lomsadze A and Borodovsky M 2001 GeneMarkS A self training method for prediction of gene starts in microbial genomes Implications for finding sequence motifs in regu latory regions Nucleic Acids Res 29 2607 2618 Borodovsky M and McIninch J 1993 GeneMark Parallel gene recognition for both DNA strands Comput Chem 17 123 133 Lukashin A V and Borodovsky M 1998 Gene Mark hmm New solutions for gene finding Nu cleic Acids Res 26 1107 1115 Pearson W R 1990 Rapid and sensitive sequence comparison with FASTP and FASTA Methods Enzymol 183 63 98 Rabiner L R 1989 A tutorial on hidden Markov models and selected applications in speech recog nition Proceedings of the LE E E 77 257 286 Internet Resources http opal biology gatech edu GeneMark GeneMark hmm Web site http opal biology gatech edu GeneMark genemarks cgi GeneMarkS Web site Contributed by Mark Borodovsky School of Biology and School of Biomedical Engineering G
30. C Gingeras T R Harrow J Hubbard T Lewis S E and Reese M G 2006 EGASP The human ENCODE Genome Annotation Assessment Project Genome Biol 7 S21 S31 Hubbard T J Aken B L Beal K Ballester B Caccamo M Chen Y Clarke L Coates G Cunningham F Cutts T Down T Dyer S C Fitzgerald S Fernandez Banet J Graf S Haider S Hammond M Herrero J Holland R Howe K Howe K Johnson N Kahari A Keefe D Kokocinski F Kulesha E Lawson D Longden I Melsopp C Megy K Meidl P Ouverdin B Parker A Prlic A Rice S Rios D Schuster M Sealy I Severin J Slater G Smedley D Spudich G Trevan ion S Vilella A Vogel J White S Wood M Cox T Curwen V Durbin R Fernandez Suarez X M Flicek P Kasprzyk A Proctor G Searle S Smith J Ureta Vidal A and Birney E 2007 Ensembl 2007 Nucleic Acids Res 35 D610 D617 Keibler E and Brent M R 2003 Eval A soft ware package for analysis of genome annota tions BMC Bioinformatics 4 50 Korf I Flicek P Duan D and Brent M R 2001 Integrating genomic homology into gene structure prediction Bioinformatics 17 S140 S148 Mouse Genome Sequencing Consortium et al 2002 Initial sequencing and comparative analy sis of the mouse genome Nature 420 520 562 Parra G Agarwal P Abril J F Wiehe T Fickett J W and Guigo R 2003 Comparativ
31. Cpg Islands Find CpG Islands using Grail1 3 I Repetitive Elements Locate repetitive elements using a BLAST based method against the Repeatmasker database DNA Sequence Raw or FASTA format paste in box or upload file a a OE ee meet Figure 4 9 1 GrailEXP submission form Select parameters 2 Select the organism of interest for analysis from the Select organism pull down menu GrailEXP and Genome Analysis Pipeline for Genome Annotation 4 9 2 Supplement 8 Currently supported organisms include human mouse Arabidopsis thaliana and Droso phila melanogaster Support for additional organisms is under development 3 Select the desired output format from the Select output type pull down menu The options for output format are Human Readable Text Genome Channel Format and Raw GrailEXP Format Human Readable Text is the only format that is designed for easy visual comprehension and is the recommended choice The other two formats are machine readable formats used by other tools at ORNL Several other output formats such as GenBank and GFF are under development 4 Check Perceval Exon Candidates checkbox if GRAIL 2 exon prediction option is desired Checking the box labeled Perceval Exon Candidates will run Perceval an updated version of the GRAIL 2 neural net based exon prediction on the input sequence This program locates exons using pattern recognition technique o
32. Download and install the appropriate MZEF executable file and all of the required data files All of the files are available by running an FTP session as follows sftp cshl org Name anonymous Password your internet address ftp gt cd pub science mzhanglab mzef ftp gt get README ftp gt cd human ftp gt binary ftp gt get mzef mzef new ftp gt mget answer yes to all the files this will download the required data files ftp gt quit The instructions on how to install MZEF are in the README file which also has a brief description about the program and parameters The command get mzef mzef_new downloads the executable mzef and renames it mzef new Change the permissions on the executable by issuing the following command chmod rwx mzef_new Run the interactive version of MZEF locally on a Unix Linux machine The results are shown in Figure 4 2 3 Current Protocols in Bioinformatics smzef new ENTER NAME OF THE SEQUENCE FILE in single quotes m12523 fasta ENTER 1 FOR FORWARD 2 FOR REVERSE il ENTER PRIOR PROBABILITY suggesting 04 04 ENTER OVER LAPPING NUMBER suggesting 0 0 See Critical Parameters for further discussion of these parameters For this example the new prior probability value Prior 0 04 was used instead of the Web default 0 02 therefore one can see some additional exon predictions in the output exon 2564 2821 was missed in Basic Protocols I and 2 becaus
33. G Agarwal P Obril J F Wiehe T Fickett J W and Guig R 2003 Comparative gene prediction in human and mouse Genome Res 13 108 117 Pearson W R 1990 Rapid and sensitive sequence comparison with FASTP and FASTA Methods Enzymol 183 63 98 Rat Genome Sequencing Project Consortium 2004 Genome sequence of the brown Norway rat yields insights into mammalian evolution Na ture 428 493 521 Stormo G D 2000 Gene finding approaches for eukaryotes Genome Res 10 394 397 Zhang M Q 2002 Computational prediction of eu karyotic protein coding genes Nat Rev Genet 3 698 709 Key References Guig6 et al 1992 See above Description of the first implementation of geneid Guig6 et al 2006 See above A community experiment to assess the state of the art in one percent of the human genome sequence Parra et al 2000 See above Description of geneid v 1 0 used in the Adh region of Drosophila melanogaster Internet Resources http genome imim es software geneid index html This is the geneid Web page http genome imim es software gfftools GFF2PS html This is gff2ps Web page http www fruitfly org annot apollo This is Apollo Web page see UNIT 9 5 http genome ucsc edu This is UCSC genome browser golden path UNIT 1 A http www sanger ac uk Software formats GFF GFF_Spec shtml This is GFF format Web page http www w3 org XML This is XML format Web page Contributed
34. GENSCAN http genes mit edu GENSCAN html Genotator http www fruitfly org nomi genotator GlimmerM UNIT 4 4 http www tigr org software glimmerm GRAIL UNIT 4 9 http compbio ornl gov tools index shtml GRAIL EXP UNIT 4 9 http compbio ornl gov grailexp HMMgene http www cbs dtu dk services HMM gene MZEF UNIT 4 2 http www cshl org genefinder PROCRUSTES http www hto usc edu software procrustes RepeatMasker UNIT 4 10 http ftp genome washington edu RM RepeatMasker html Sputnik http rast abajian com sputnik http genes cse wustl edu An Overview of Gene Identification 4 1 2 Supplement 6 promoter gt exon 1 intron1 exon2 intron2 exon3 intron 3 exon 4 transcription RNA cap 5 eng modification polyA y GU AG GU AG GU_AG man spicing stop mature fs mRNA P poya nucleus cytoplasm translation 2 2 polyA Figure 4 1 1 The central dogma of molecular biology Proceeding from the DNA through the RNA to the protein level various sequence features and modifications can be identified that can be used in the computational deduction of gene structure These include the presence of promoter and regulatory regions intron exon boundaries and both start and stop signals Unfortunately these signals are not always present and when present may not always be in the
35. Genes 4 6 5 Supplement 1 Eukaryotic Gene Prediction Using GeneMark hmm 4 6 6 Supplement 1 gt gi 178344 gb AAA98797 1 albumin 609 aa MKWVTFISLLFLFSSAYSRGVFRRDAHKSEVAHRFKDLGEENFKALVLIA FAQYLQQCPFEDHVKLVNEVTEFAKTCVADESAENCDKSLHTLFGDKLCT VATLRETYGEMADCCAKQEPERNECFLQHKDDNPNLPRLVRPEVDVMCTA FHDNEETFLKKYLYEIARRHPYFYAPELLFFAKRYKAAFTECCQAADKAA CLLPKLDELRDEGKASSAKQRLKCASLOKFGERAFKAWAVARLSQRF PKA EFAEVSKLVTDLTKVHTECCHGDLLECADDRADLAKYICENQDSISSKLK ECCEKPLLEKSHCIAEVENDEMPADLPSLAADFVESKDVCKNYAEAKDVF LGMFLYEYARRHPDYSVVLLLRLAKTYETTLEKCCAAADPHECYAKVFDE FKPLVEEPQNLI KQNCELFEQLGEYKFQNALLVRYTKKVPQVSTPTLVEV SRNLGKVGSKCCKHPEAKRMPCAEDYLSVVLNQLCVLHEKTPVSDRVTKC CTESLVNRRPCFSALEVDETYVPKEFNAETFTFHADICTLSEKERQIKKQ TALVELVKHKPKATKEQLKAVMDDFAAFVEKCCKADDKETCFAEEGKKLV AASQAALGL gt gi 178345 gb AAA98798 1 alloalbumin Venezia 604 bases MKWVTFISLLFLFSSAYSRGVFRRDAHKSEVAHRFKDLGEENFKALVLIA FAQYLQQCPFEDHVKLVNEVTEFAKTCVADESAENCDKSLHTLFGDKLCT VATLRETYGEMADCCAKQEPERNECFLQOHKDDNPNLPRLVRPEVDVMCTA FHDNEETFLKKYLYEIARRHPYFYAPELLFFAKRYKAAFTECCQAADKAA CLLPKLDELRDEGKASSAKQRLKCASLOKFGERAFKAWAVARLSQRF PKA EFAEVSKLVTDLTKVHTECCHGDLLECADDRADLAKYICENQDSISSKLK ECCEKPLLEKSHCIAEVENDEMPADLPSLAADFVESKDVCKNYAEAKDVF LGMFLYEYARRHPDYSVVLLLRLAKTYETTLEKCCAAADPHECYAKVFDE FKPLVEEPQNLI KQNCELFEQLGEYKFQNALLVRYTKKVPQVSTPTLVEV SRNLGKVGSKCCKHPEAKRMPCAEDYLSVVLNQLCVLHEKTPVSDRVTKC CTESLVNRRPCFSALEVDETYVPKEFNAETFTFHADICTLSEKERQIKKQ TALVELVKHKPKATKEQLKAVMDDFAAFVEKCCKADDKETCFAEEPTMRI
36. It may be accessed at the AAT Analysis and Annotational Tool Web site http genome cs mtu edu aat aat html If one uses the same sequence and parameters as the example of the interactive MZEF run see Alternate Protocol one will obtain the results shown in Figure 4 2 4 from the AAT server It can be seen that the two false positive internal coding exons 13341 13425 and 17812 17874 have been eliminated due to the lack of EST matches There is a danger that a novel exon may also be eliminated Another related program is called MZEF SPC Thanaraj and Robinson 2000 which is an integrated system for exon finding with SpliceProximalCheck as a front end for MZEF It may be accessed at the EBI Web site http industry ebi ac uk thanaraj MZEF SPC html If one uses the same sequence and parameters as the example of the command line MZEF run see Figure 4 2 2 one will obtain the results shown in Figure 4 2 5 from the MZEF SPC server Since Overlap was set to 1 the default Overlap 10 in the MZEF SPC server among overlapping MZEF predicted exons MZEF SPC was able to pick out most of the exons correctly except the last one On average however MZEF SPC should pick out more true exons among overlapping ones than MZEF nonoverlapping predictions When selecting true exons among possible ones frame compatibility should also be considered Critical Parameters and Troubleshooting As mentioned above MZFF requires three input parameters oth
37. MIRbDSSINE MIR 176 TIGITGIGAGGATTAAATGAGATAATGCATGTAAAGCGCTTAGCACAGTG 225 hgi6_dna 662 CITTTCATAAGGTAAGCACTTIGAAAATATICACTTITACTATT 705 ivy ivi i iviv Svan 2 o MIRbSSINE MIR 226 CCTGGCACACAGTAAGCGCTCAATAAATGGIAGCICI ATTATT 268 Figure 4 10 5 Alignments between query sequence and consensus repetitive elements are shown if the option BASIC PROTOCOL 2 Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences 4 10 6 Supplement 25 Show Alignments is selected 8 In the series of pull down menus under Advanced Options select the appropriate Options These options are straightforward as well For example if the user wants to make a choice between the Masking Options users can either choose ambiguous characters like N or X for masking or lowercase letters which may be more appropriate for subsequent alignments Detailed explanation of these and additional options available can be accessed by clicking on the link to the right of each pull down menu 9 Click the Submit Sequence button to run RepeatMasker The results displayed in the browser are shown in Figures 4 10 2 4 10 3 4 10 4 and 4 10 5 See Guidelines for Understanding Results for details USING THE COMMAND LINE Unix Linux VERSION OF RepeatMasker TO STUDY REPETITIVE ELEMENTS IN GENOMIC SEQUENCES Command line RepeatMasker provides users with more choices and does not have the 100 kb length limit for query sequences To run Repe
38. RERK Figure 4 6 5 The annotated sequences for the two proteins resulting from two variants of alternative splicing in the last exon of human serum albumin are shown The amino acid residues not included in the prediction Fig 4 6 4 are shown in bold may help to identify additional candidates for alternative splicing which GeneMark hmm alone would not detect as it only predicts a single structure for each gene The regions with high coding potential can be reported by GeneMark in text format as described in UNIT 4 5 Figure 4 6 6 depicts the coding potential in the six possible reading frames three frames on the direct strand and three frames on the reverse strand An unbroken horizontal line at the 0 5 level indicates an open reading frame ORF Vertical ticks under the 0 5 level represent one of the three stop codons TAA TGA or TAG The thick gray horizontal line indicates a region with higher than expected coding potential predicted by GeneMark The lt and gt marks represent putative exon boundaries acceptor and donor splice sites respectively within a region of high coding potential The thick black horizontal lines at the bottom level of each panel indicate exons predicted by GeneMark hmm Eukaryotic GeneMark hmm presents the graphical output in Adobe PostScript format which is sent by E mail to the user Some E mail readers will allow the figure to be displayed automatically while others may requ
39. The param eter files for different species have been devel oped by Gen s Parra and Francisco Camara The current Web server has been written by Enrique Blanco at the IMIM geneid uses Position Weight Matrices PWM to predict potential splice sites and start codons Potential sites are scored as log likelihood ratios From the set of predicted sites which includes in addition all potential stop codons the set is built of all potential ex ons Exons are scored as the sum of the scores of the defining sites plus the log likelihood ra tio of the Markov model for coding sequences Finally the gene structure is assembled from the set of predicted exons maximizing the sum of the scores of the assembled exons Predicting and scoring sites PWMs are used to score each potential donor site GT acceptor site AG and start codon ATG along a given sequence The score of a potential donor site if assumed to be of length S s s2 s within the sequence is computed as I Ly 8 Dey j l This is the log likelihood ratio of the sequence S in an actual site versus S in any false GT site Dj is the logarithm of the ratio of the proba bility of nucleotide i in position j in an actual donor site over the probability of 7 in position jin a false site D j values are estimated from a training set of positive and false donor sites Similar scores are computed for Acceptor Sites La and Start Codons Lg Predicting and
40. Unix application The best way to take full advantage of the different options available in geneid is by running the stand alone program on a Unix workstation In both cases the user provides an input DNA sequence as a FASTA file APPENDIX 1B and selects a suitable model of parameters depending on the species or taxonomic group from which the sequence originates A number of options are available to configure geneid actions and output Although this option is not directly available in the stand alone Unix application geneid output can be directly plugged into a number of publicly available visualization tools see Basic Protocol 2 This protocol describes the use of geneid as a stand alone Unix application For use of the geneid Web server as an alternative see Alternate Protocol Necessary Resources Hardware Unix Linux workstation with at least 256 Mb RAM recommended Contributed by Enrique Blanco Gen s Parra and Roderic Guig Current Protocols in Bioinformatics 2007 4 3 1 4 3 28 Copyright 2007 by John Wiley amp Sons Inc UNIT 4 3 BASIC PROTOCOL 1 Finding Genes 4 3 1 Supplement 18 Using geneid to Identify Genes 4 3 2 Supplement 18 Software geneid v1 2 full distribution see Support Protocol Files This protocol uses the following file examplel fa 32 kb masked It is a human genomic sequence extracted from the UCSC human genome browser assembly March 2006 location huma
41. additional convenience of GeneMark hmm is the simultaneous running of Typical and Atypical models Typical models are built from the major class of genes and Atypical models are built from the minor class presumably laterally transferred genes GeneMark see Basic Protocol 1 can optionally be run simulta neously with GeneMark hmm for these models as well The output of the analysis can be presented in text Fig 4 5 5 and graphical Fig 4 5 6 formats The Typical models that are used with GeneMark hmm have been constructed using the GeneMarkS program see Alternate Protocol 2 The Atypical models have been built using the Heuristic model approach see Basic Protocol 2 The Typical model is sensitive to the genes of the mainstream population sharing the same codon usage pattern The Atypical model can find genes that do not match this common pattern which often occurs with genes that have been horizontally transferred Both models used in parallel find the vast majority of genes in a particular genome Necessary Resources Hardware A personal computer or workstation with Web access Software A Web browser Current Protocols in Bioinformatics F GeneMark hmm for Prokaryetes Microsoft Internet Explorer fhe Cet Yew fpote Tock tee Ou O 2 Q Ps Syne Qua O M L Me ios areg geh espandir itan prot og GeneMark hmm for Prokaryotes and Low Eukaryotes Version 2 1 ort s psp Reference Lukeshin A and Borodovsky M GeneMark hmm
42. alternatively spliced is not predicted Prediction of initial and terminal exons is usually problematic in all gene prediction programs due to the rather weak statistical patterns of the translation start and stop Figure 4 6 4 displays the predicted sequence of human serum albumin ALB The anno tated sequences for the two proteins resulting from two variants of alternative splicing in the last exon are shown in Figure 4 6 5 The amino acid residues not included in the prediction are shown in bold in the annotated sequence 6 Interpret the graphical output The GeneMark hmm graphical output is always combined with the GeneMark graphical output Fig 4 6 6 also see UNIT 4 5 Basic Protocol Graphical output in the PostScript format is generated by the GeneMark program using the models relevant to the species in question The file is generated and E mailed to the user if the Generate PostScript Graphics box was checked see step 3 In this case GeneMark runs concurrently with the eukaryotic GeneMark hmm program While primarily used for prokaryotic DNA analysis see UNIT 4 5 GeneMark can also be used to aid eukaryotic gene finding The GeneMark graphical output Fig 4 6 6 identifies regions with high coding potential both inside and outside of open reading frames These regions can be used together with the exon predictions of GeneMark hmm to further analyze the sequence Particularly it Current Protocols in Bioinformatics Finding
43. bank recent devel opments Nucl Acids Res 21 3093 3094 Beck S Kelly A Radley E Khurshid F Alder ton R P and Trowsdale J 1992 DNA sequence analysis of 66 kb of the human MHC class II region encoding a cluster of genes for antigen processing J Mol Biol 228 433 441 Benson D Lipman D J and Ostell J 1993 Gen Bank Nucl Acids Res 21 2963 2965 Current Protocols in Bioinformatics Bilofsky H S and Burks C 1988 The GenBank genetic sequence data bank Nucl Acids Res 16 1861 1864 Boguski M S Lowe T M and Tolstoshev C M 1993 dbEST database for expressed se quence tags Nature Genet 4 332 333 Brody L C Abel K J Castilla L H Couch F J McKinley D R Yin G Y Ho P P Merajver S Chandrasekharappa S C Xu J Cole J L Struewing J P Valdes J M Collins F S and Weber B L 1995 Construction of a transcrip tion map surrounding the BRCA1 locus of hu man chromosome 17 Genomics 26 238 247 Fields C Adams M D White O and Venter J C 1994 How many genes in the human genome Nature Genet 7 345 346 Gardiner Garden M and Frommer M 1987 CpG islands in vertebrate genomes J Mol Biol 196 261 282 John R M Robbins C A and Myers R M 1994 Identification of genes within CpG enriched DNA from human chromosome 4p 16 3 Human Mol Gen 3 1611 1616 Jurka J Walichiewicz J and Milosavljevic A 1992 Prototypic sequences
44. be cause there is additional information that can help with the alignment process most notably that the internal edges of each alignment piece should fall on standard splice site GT AG boundaries A gene message align ment program aligns a gene message with a genomic sequence such that the exon intron boundaries are clearly identified The databases currently used by Galahad at ORNL are CBIL Penn DOTS Database TIGR EGAD Transcript Database NCBI RefSeq Da tabase NCBI dbEST Database and a curated set of mRNAs from Genbank The default input to Galahad is a DNA query sequence and a list of GRAIL Exon Candidates It replaces all non exonic portions of the query sequence with N s and runs a BLAST search of this sequence against the search database Only the high scoring BLAST alignments are retained thus eliminating a significant number of poor alignments A temporary database is created from the retained ESTs cDNAs and a BLAST search unr 3 3 of the entire sequence is run against this database These raw BLAST alignments are arranged into valid gene struc tures by applying a dynamic programming al gorithm A merge check algorithm locates the fragments that BLAST failed to join and merges these alignments An exhaustive search of all splice sites in the vicinity of each edge is per formed in order to select the optimal set of left right and internal edges A donor splice site scoring algorithm is used to pick the mo
45. by Enrique Blanco Genis Parra and Roderic Guig Centre de Regulaci Gen mica Institut Municipal d Investigaci M dica Universitat Pompeu Fabra Barcelona Spain Current Protocols in Bioinformatics Using GlimmerM to Find Genes in UNIT 4 4 Eukaryotic Genomes GlimmerM is a gene finder originally developed for small eukaryotes particularly for organisms with a relatively high gene density Salzberg et al 1999 The original system was designed to find genes in Plasmodium falciparum the malaria parasite Gardner et al 1998 With the demands of many recent genome sequencing projects each calling for its own gene finder the system has been trained for many additional organisms including Arabidopsis thaliana Oryza sativa Yuan et al 2001 Theileria parva and Aspergillus fumigatus It performs well on all of these even those with relatively low gene density and on closely related organisms A special package included with the latest release of GlimmerM re trains the system using data provided by the user thereby making the gene finder applicable to virtually any organism limited only by the availability of training data Information on how to obtain the Unix version of GlimmerM software is presented in the Basic Protocol This section also describes the usage of the system to predict gene models in genomic DNA sequences The Support Protocol presents the steps required by the training procedure of GlimmerM Early versions
46. column may contain Initial Internal or Terminal terms describing the type of predicted exon For single exon genes Exon Type is assigned as Single The list of predicted exons shows start and end positions of each exon in the Exon range columns The Start End Frame columns specify the positions of the first and the last nucleotide of an exon in terms of codon position If the Translate Predicted Genes into Proteins box was checked see step 3 the sequences of predicted proteins will be displayed in FASTA format in the output below the list of predicted genes and exons Fig 4 6 4 Current Protocols in Bioinformatics fe Ede Yew apota Too Hp Mp ep aLneingy gatech IAA eum Y GeneMark heem Protein Translations Ge to GeneMark bmm Listiog Go to Job Sebmittal EL VEER PRKATKEQLFAICODEAAFVERCCKADOKETCFALEL Input Sequence Title optional Sequence e Sequence File upload e Browse Figure 4 6 4 The text output the Eukaryotic GeneMark hmm program containing the sequence of predicted protein in FASTA format Figure 4 6 2 displays the results for gene prediction using the sample sequence for human serum albumin ALB gene M12523 Thirteen of fourteen annotated exons are predicted and twelve of the predicted exons are predicted exactly matching the start and the end of each exon The predicted start position of the initial exon does not match the annotation and the last exon known to be
47. determined Such coordinates of the promoter element are given in the GFF file samples example3 promoter gff example3 experimental Promoter 1500 1799 4 The prediction includes now a first coding exon similar to the annotated one Fig 4 3 9E Current Protocols in Bioinformatics USING THE geneid WEB SERVER TO PREDICT GENES A Web interface to geneid can be accessed at http genome imim es geneid html The geneid server consists of a form to input the DNA sequence which is mandatory as well as the external information to improve the prediction which is optional providing a set of different options to customize the behavior of the program All of the geneid functionality is available through the geneid Web server in particular the operations and commands described in the previous protocols see Basic Protocol 1 and 3 Moreover this server can supply a graphical representation of the predictions obtained with the program gff2ps see Basic Protocol 2 This protocol outlines the use of this interface to predict genes as well as other genomic elements on DNA sequences The geneid Web server is divided basically into three main areas according to the type of information they provide to the user Input Data Fig 4 3 10 Prediction Options Fig 4 3 11 and Output options Fig 4 3 12 Once the user has supplied a sequence to process and selected the appropriate options the form containing this information must be transferred f
48. es Suggestions for Further Analysis The authors are investigating a number of extensions to geneid which are not discussed above 1 Incorporating homology information into the gene predictions For instance such information can be obtained after the compari son of the query sequence against a database of known amino acid sequences using BLASTX Altschul et al 1990 UNIT 3 4 or FASTA Pearson 1990 Processed database search re sults can already be passed to geneid via S option The authors have chosen here not to discuss this option because the use of homol ogy information requires fine tuning of some of the geneid parameters tuning that the au thors have not performed yet Still the option S can be of utility For instance in Basic Pro tocol 3 when passing to geneid the coordinates of EST fragments via the R option these are processed as corresponding exactly to coding exons Often however the exact coordinates of an exon are not known for instance when matching similar but not identical ESTs or when the EST expands into the UTR In such a case the coordinates of the region in which the exons are suspected can be given to geneid via the S option geneid then will rescore all candidate exons overlapping the region The resulting exon score will be a function of the original exon score the score of the region and the degree of overlap between the region and the exon If the score given to the region is
49. et al 1998 C trachomatis Stephens et al 1998 T maritima Nelson et al 2001 V cholerae Heidelberg et al 2000 and many other prokaryotes Based on the success of Glimmer in bacterial sequence annotation the authors hypothesized that IMMs would make a good foundation for eukaryotic gene finding This is particularly true of small eukaryotes like P falciparum in which the gene density is inter mediate between that of prokaryotes and higher eukaryotes To predict genes in malaria GlimmerM runs separately over both the direct and complemen tary strands of the input The algorithm then makes one more pass over the list of putative genes to reject overlapping genes If genes overlap by less than a fixed amount 30 bp by default then the overlap is ignored and both genes are reported in the output Most overlap ping genes are competing gene models that share a stop codon with one or more alternative exons comprising the only differences between the models Genes that overlap by gt 30 bp are re scored using the IMM and the one with the best score is retained If the scores of two or more overlapping models differ from the maxi mum score by less than a small pre set amount then GlimmerM considers the scores equiva lent and outputs all the models as possible genes In these instances it marks the longest Current Protocols in Bioinformatics gene as the preferred one Of course the over lapping gene models predic
50. exon intron and gene boundaries The previously developed program GeneMark Borodovsky and McIninch 1993 identified a gene mainly as the open reading frame where the gene resides However the Web site ver sion of the GeneMark program does not use a notion of exons and introns Therefore it does not predict exon intron boundaries as such The underlying idea of GeneMark hmm was to embed the GeneMark models for pro tein coding exons and non coding intron and intergenic regions into a naturally de rived hidden Markov model HMM frame work with exon intron boundaries modeled as transitions between hidden states Current Protocols in Bioinformatics The HMM framework of GeneMark hmm the logic of transitions between hidden Markov states followed the logic of the genetic struc ture of eukaryotic genome The Markov models of coding and non coding regions were incor porated into the HMM framework to generate stretches of DNA sequence with coding or noncoding statistical patterns This type of HMM architecture is known as HMM with duration Rabiner 1989 The sequence of hidden states associated with a given DNA sequence carries information on positions where coding function is switching into non coding and vice versa The sequence of hidden states constitutes the HMM trajectory The core GeneMark hmm procedure the dynamic pro gramming type algorithm Rabiner 1989 finds the most likely HMM trajectory given the DNA sequ
51. explanation are shown in Figure 4 3 13 The input file DNA sequence was given in the Input Data section see Fig 4 3 10 and the option Do You Want a Graphical Representation of the Predictions in the Input Data section was checked In the Prediction Options section see Fig 4 3 11 Homo sapiens was selected for Organism Normal mode was Selected for Prediction Mode and Forward and Reverse was selected for DNA Strands In the Output Options section GFF was selected for Output Format Finding Genes 4 3 17 Current Protocols in Bioinformatics Supplement 18 SUPPORT PROTOCOL Using geneid to Identify Genes 4 3 18 Supplement 18 geneid predictions on sequence submitted from rantamplan bio ub es are 8 gff version 2 date Mon Jan 22 17 45 40 2007 source version geneid v 1 2 geneid imim es Sequence examplel Length 32001 bps Optimal Gene Structure 1 genes Score 16 70 Gene 1 Forward 8 exons 470 aa Score 16 70 examplel geneid_v1 2 First 736 1130 6 14 o examplel_1 examplel geneid_v1 2 Internal 5504 5618 0 49 1 examplel_1 examplel geneid_v1 2 Internal 5778 5951 1 13 o examplel_1 examplel geneid_vl 2 Internal 8730 8836 0 84 o examplel_1 examplel geneid_v1 2 Internal 13186 13256 0 46 1 examplel_1 examplel geneid_v1 2 Internal 21287 21488 2 78 2 examplel_1 examplel geneid_vl 2 Internal 29896 30019 1 56 1 examplel_1l examplel geneid v1 2 Terminal 31726 31
52. external information in the geneid prediction by using the option R geneid P param human3iso param R samples example2 evidences gff samples example2 fa Gene features exons and genes can be externally provided to geneid The program then produces gene predictions that incorporate these features These gene features are supplied in a GFF file External gene features must be of a geneid exon type First Internal Terminal or Single to work with partially supported exons see Suggestions for Further Analysis The strand on which they occur must also be provided but frame and score are optional by placing a in the GFF corresponding field The GFF fields seqname and source are not used and they can be anything Users should be aware however that if a score is specified for provided exons these will compete with geneid predicted exons and may not be included in the final prediction The group field in the GFF file can be used to prevent geneid from predicting additional exons within a known gene Exons with the same group identifier are considered to belong to the same gene and no additional exon is predicted between them see geneid manual for details External gene features are provided to geneid by means of the R option followed by the name of the GFF file In the case of the example the GFF file including the exon coordinates of the known gene is remember that in GFF fields are delimited by tabs 13800 27600 A
53. formats for other available options refer to repeatmasker help a shows the alignments in a align output file small returns complete masked sequence in lower case xsmall returns repetitive regions in lowercase rest capitals rather than masked X returns repetitive regions masked with Xs rather than Ns gff creates an additional General Feature Finding format output Note that the cut option is not supported in the current release of RepeatMasker however the function may be obtained by contacting Robert Hubley rhubley systemsbiology org RUNNING REPEATMASKER WITH WU BLAST Running RepeatMasker for larger sequences e g whole genome for Homo sapiens will take a significant amount of time The processing time can be reduced roughly 30 fold by using WU BLAST as the engine for RepeatMasker to replace cross_match Bedell et al 2000 Although RepeatMasker with WU BLAST has better processing time the combination also has some limitations 1 low complexity repeats are not as efficiently masked as when RepeatMasker is used with cross_match 2 some output formats are not supported and 3 the accuracy of the results returned by the combination of RepeatMasker with WU BLAST has not been assessed NOTE Investigators unfamiliar with the Unix environment should read APPENDIX 1C and APPENDIX 1D Necessary Resources Hardware Unix or Linux workstation Software RepeatMasker see Basic Protocol 2 WU BLAST 2 0 co
54. from human repeti tive DNA J Mol Evol 35 286 291 Larsen F Gundersen G Lopez R and Prydz H 1992 CpG islands as gene markers in the human genome Genomics 13 1095 1107 Lawrence B J Schwabe W Kloschis P Coy J F Poustka A Brennan M B and Hochgesch wender U 1994 Rapid identification of gene sequences for transcriptional map assembly by direct cDNA screening of genomic reference libraries Hum Mol Gen 3 2014 2023 Marshall E 1995 A strategy for sequencing the genome 5 years early Science 267 783 784 Current Protocols in Bioinformatics Pearson W R and Lipman D J 1988 Improved tools for biological sequence comparison Proc Natl Acad Sci U S A 85 2444 2448 Peltoketo H Isomaa V Maeentausta O and Vi hko R 1988 Complete amino acid sequence of human placenta 17 B hydroxysteroid dehydro genase deduced from cDNA FEBS Lett 239 73 77 Smith T F and Waterman M S 1981 Identification of common molecular subsequences J Mol Biol 147 195 197 Smith M W Holmsen A L Wei Y H Peterson M and Evans G A 1994 Genomic sequence sampling A strategy for high resolution se quence based physical mapping of complex genomes Nature Genet 6 40 47 Wiginton D A Kaplan D States J C Akeson A L Perme C M Bilyk I J Vaughn A J Lattier D C and Hutton J J 1986 Complete sequence and structure of the gene for human adenosine deaminase Biochemi
55. g Send LINE Li 1885 216 1 hg10_dna 3 GGCAAATACCTCAAAGGAAAAATTG CTTAAATGTTAGACTTA TCTA 48 i aa ad vi 2444 i m p LIMEg_Send LI 216 GGCAAATGCCTCGAGGGGAAAAGCGGCHCCGAATGICGGGCICACCICTC 167 hg18_dna 49 GGAACTTIGITICCCTCTGGAATCTIGGCCCTIGAAGC CITIGCTAATT 97 v i ava a i i av iitii ivi LIMEg_Send LI 166 TGGGCTICCCTICTCICCGGGATCTITGGCCCCTCAAATICCTCGCTGCCT 117 hgi6_dna 98 TGGTAAGCTTTCTGATGICTTCAAACAAGTATTITITAAAACATITICIGGA 147 i i i ii i vy i vv Lid g_Send LI 116 TIGGTA GCTCTCCGATGCCTICAAACAGATGITITITA TATITIGICCA 69 hg10_dna 248 G TITTATAATTGICCTTGGCAGCAGTGCTGGTTTGTAATAGGCTTATCT 195 wv i ii iv vi i 64 vi LIMEg_Send LI 8 GCTITITCTAGITTGITCTCAGCGGGAGGGTTGGTCTG AACAANCTAGTCC 20 hg18_dna 196 ATCATTACTGGAAGIGGAA 214 ai F gt a LikEg_Send LI 19 GCCATTGCCGGAAGCGGAA 1 359 32 68 12 96 0 83 hgif dna 490 705 20439 MIRb SINE MIR 27 268 0 hg10_dna 490 AAAGGAGGGATATTGAAAT AGTATGAGTTGGAGTCAACTCTTGATTTC 537 i vv E ek er Teh av iw ai da i8 MIRDSSINE MIR 27 AGAGCACGGGCTTTGGAGTCAGGCA GACCIGGGITCGAATCCIGGCICT 75 hgi8_dna 38 ACAACTTACT GIGIC TCTTTTTCAACTIGITIAACCTCITTATG 81 iv oo v i vvv v iii T MIRb SINE MIR 76 GCCACTTACTAGCTGTGTGACCTTGGGCAAGTCACTTAACCTCTCTGAGC 125 hgi8_dna 82 TCAGITTCICCATCA AAAATATGGA AGTAATA TAAAT AGGA 622 aa g iv a qp ame i MIRb SINE MIR 126 CTCAGTTTCCTCATCTGTAAAATGGGGATAATAATACCTACCTOGCAGGG 175 hg10_dna 23 TTATTGIGAGTGTTAAATAAGGTCATGTATATAAA 99 AATG 61 a vi iv i See a
56. gt 2 In order to distinguish one class object from another one needs two things a set of feature variables x x y Q 1 p and a decision rule i e classifier C such that given the measured values x for the ith object C would be able to map it into either class I denoted by or class II denoted by see Figure 4 2 6 In practice choosing the set of feature variables that is most discriminative with respect to the two classes is the key to success For example sex hormone level is a much better discriminative feature variable than weight when classifying people as males and females Although there are many systematic methods for selecting better feature variables it is still more or less like a black art which depends heavily on the master s insight to the nature of the subject Once the set of feature Xk Figure 4 2 6 A classifier C separates N 13 sample points in K 2 feature space Error 1 Current Protocols in Bioinformatics variables is decided or given one can represent the N objects to be classified as N sample points x in the p dimensional feature space Discriminant theory will offer the mathe matical tools for finding the optimal classifier in the sense of minimizing the classification errors In general the Bayesian theory assumes the sample points were drawn from two distinct distributions p xl f x and p xl f x If these conditional distributions and
57. gt DD swch ife Guda J D J Addons E ip nde cti cag cg tantoak F eu F Art cg FentEF Rens Sun Sep 29 13 01 01 2002 uploaded sequence Ge chr20300000 400000 na gt chr20 300001 400000 1 Predictions om the direct strand No Promoter P promoter Exon Piexon Pidonor CpG Vindow Pank 2 00016397 00016966 1 0000 00016097 00017093 1 0000 0 9904 00015748 00015949 i 3 00015220 00015797 1 0000 00015720 00015043 1 0000 0 9970 00015748 000159419 2 1 00016397 00016966 1 0000 O00016897 00017296 1 0000 0 9962 00016402 00016403 3 1 00016397 00016966 1 0000 00016897 00017242 1 0000 0 9952 00015748 00015949 4 2 00021303 00021952 0 6619 21083 00022001 0 5965 0 9642 Non CpG related 3 2 00049534 00050103 1 0000 00050034 0005023 1 0000 0 99708 000490932 000500 a 3 00049534 00050103 1 0000 00050034 00050359 1 0000 0 9987 00049632 00050033 2 3 00049534 00050103 1 0000 O0050034 00050321 1 0000 0 9967 00049832 00050033 3 3 00049475 00050044 1 0000 00049975 00050040 1 0000 0 9965 00049632 00050033 4 3 00049534 00050103 1 0000 00050034 00080201 1 0000 0 9672 00049632 00050033 3 00049534 00080103 1 0000 00050034 00050327 1 0000 0 9717 000419632 00050033 s 3 00049531 00050100 1 0000 0005003 1 00050092 1 0000 0 9682 00049632 00050033 7 3 00049273 00049042 1 0000 00049773 00049070 1 0000 0 9553 00049632 00050033 0 4 00076637 00077206 1 0000 O0077137 00077480 1 0000 0 9268 00076715 00076916 i S 0006915
58. iden tify genes or the protein coding portions of genes by identifying specific functional sites such as splice donor and splice acceptor junc tions Brunak et al 1990 1992 Guigo et al 1992 Hutchinson and Hayden 1992 Tech niques have also been devised to locate coding regions by their statistical Claverie et al 1990 or periodic properties Mani 1992 In addi tion there are gene finding techniques that combine these types of information in various ways Gelfand 1990 Uberbacher and Mural 1991 Mural et al 1992 Snyder and Stormo 1993 Xu et al 1994a The first GRAIL system GRAIL 1 evalu ated seven statistical parameters distinguishing coding from noncoding regions for a window of 100 bases of sequence The values for each window were then processed by an artificial neural network that had been trained to distin guish coding from noncoding regions on the basis of these parameters This system per formed quite well it was able to recognize 90 of annotated protein coding regions 2100 bases in length from a test set of 19 genes Uberbacher and Mural 1991 An acceptably low false positive rate approximately one in five predictions when evaluated for both strands was obtained Though GRAIL 1 was very useful and widely used it left several problems unsolved The fixed window size 100 bases made it difficult for the system to recognize exons lt 100 bases in length Further more the system did
59. in the UCSC genome browser annotations It is very important to maintain the tabular separation between columns The user can select different displaying options e g color shape etc The number and type of tracks to be displayed in the image can also be selected here adding other commands in the header Press the Go to the genome browser button to show the geneid GFF predictions incorporated into the UCSC genome browser annotations The main UCSC genome window will appear again The geneid track is now displayed on top of the picture in green Coding exons are represented as boxes Exons belonging to the same gene are joined together by a line Users can easily add or remove other sources of information Figure 4 3 7 shows the UCSC display of the geneid prediction obtained in Current Protocols in Bioinformatics Finding Genes 4 3 9 Supplement 18 chr21 13925000 13935000 13910000 13915000 geneid predictions example UCSC Known Genes Based on UniProt RefSeq and GenBank mRNA Known Genes Raper i by RepeatMasker RepeatMasker Figure 4 3 7 Using the UCSC genome browser to visualize geneid output BASIC PROTOCOL 3 Using geneid to Identify Genes 4 3 10 Supplement 18 Basic Protocol 1 step 6 the Known Genes track and the Repeat Masker results see the Guidelines for Understanding Results for further discussion on gene prediction accuracy and sequence mask
60. include the option of using an RBS ribosome binding site model to aid in more accurate gene start prediction In either method the number of genes accurately predicted depends on how well the sequence is described by the model Models built from experimentally verified coding and noncoding regions are more reliable however the number of experimentally verified genes remains small and thus other methods of building accurate models have been developed The Heuristic Approach see Basic Protocol 2 builds a fairly accurate inhomogeneous Markov model of protein coding regions based on the relationships between the positional nucleotide frequencies and the global nucleotide frequencies observed in the analysis of 17 complete bacterial genomes This method can be used for sequences as small as 10 kbp GeneMarkS see Alternate Protocol 2 utilizes a nonsuper vised training procedure and can be used for a newly sequenced prokaryotic genome with no prior knowledge of any protein or rRNA genes However it requires at least 1 Mbp of sequence Models built from either of these procedures can then be used by GeneMark and GeneMark hmm for sequence analysis The programs accessible through the Web site are periodically updated Therefore the users of the Web site always have access to the latest versions of the programs USING GeneMark FOR PROKARYOTIC GENE PREDICTION The GeneMark program Web site Fig 4 5 1 uses precomputed statistical models for 32 s
61. itself 3 Enter the E mail address to which the results are to be sent optional If an E mail address is provided the server will forward the results in the body of an E mail message 4 Click Submit The server displays the results in the browser Output of FirstEF X Analyze the resulting output Fig 4 7 2 which consists of eight columns described in Table 4 7 1 and presented in two parts predictions for direct and complementary strands for boundaries of the promoter and first exon from the Promoter and Exon columns respectively For Example 1 FirstEF predicted five clusters on the direct strand and four clusters on the complementary strand The predictions within a cluster are considered as probable alternative first exons of the same gene and ranked according to a posteriori probabilities The cluster numbers of the prediction are displayed in the first column No and ranks in the last column Rank The prediction with rank 1 in cluster number 1 has a promoter spanning from 16397 to 16966 of length 570 nt and exon spanning from 16897 to 17093 Note that the first 500 nucleotides of 16397 to 16966 is upstream of the region of the predicted TSS and the last 70 nucleotides are the downstream of the predicted TSS that overlaps with the predicted first exon i e 16897 to 17093 Further this particular prediction is CpG related and hence the predicted CpG window 15748 to 15949 of length 201 is reported in CpG win
62. length for filtering locally maximal donor sites default 60 Current Protocols in Bioinformatics CJ TSO predator mpertea train trainGlimmerM data s fasta data exons 7 RSET gl SP predator mpertea Malaria train traini1_01 1 gt s TrainG imal2001 11 09914 34 57 TrainGlimmM2001 11 09D14 34 57 log mess predator apertea Halaria train traini1_01 Figure 4 4 1 Training GlimmerM for a malaria data set where the DNA sequences for training are in the file seqs fasta and the exon coordinates are in the file exons coord shown on the first and second lines By adding the text gt amp messages at the end of the trainGlimmerM command the authors created a file called messages which captured any potential messages of the training program that were normally printed on the screen The computer prompt shown is predator mpertea Malaria train trainll1_01 The third line shows the execution of the 1s command APPENDIX 1C which lists the contents of the directory fourth line The directory now contains the subdirectory TrainGlimmM2001 11 09D14 34 57 and the two files TrainGlimmM2001 11 09D14 34 57 1log and messages If insufficient data is available for training the splice sites the training procedure will be unsuccessful and exit with a warning message The system determines dynamically whether the data is sufficient by estimating error rates on both donor and acceptor sites If it fails then the user sh
63. models Columns 2 to 7 are the scores for the coding region in each of the six reading frames F1 F2 and F3 represent the three forward reading frames while F4 F5 and F6 are the reverse complement frames The score represents a normalized probability that the coding module the IMM embedded in GlimmerM generated the DNA sequence of the gene in that frame for further details see Salzberg et al 1998a The IndScore column is the normalized probability that the DNA sequence was generated by a model of independent random probabilities for each base Salzberg et al 1998a This is a simple model estimating the probability that the sequence is just random DNA with the same underlying GC composition The Splice site scores column shows the scores of the donor and acceptor splice sites of the potential gene models Salzberg 1997 Gene models with one intron have two scores shown donor and acceptor in order Models with more introns have two scores shown per intron Higher scores for the splice sites are better The last section of the output in Figure 4 4 8 begins with the phrase Putative genes and shows the gene models themselves Each model begins with an ID number followed by a list of exon positions beginning at the start codon and ending at the stop codon Note that noncoding exons are not predicted at all and that only the coding portions of the initial and terminal exons are shown Thus the initial exon begins with the trans
64. nonhuman mammal target Brassica oleracea for an Arabidopsis target Arabidopsis thaliana for a non Arabidopsis dicot plant target C briggsae for a C elegans target C elegans for a target that is a member of the genus Caenorhab ditis other than C elegans Drosophila ananassae for Drosophila melanogaster and Drosophila melanogaster for all other insects If N SCAN annotations for the target species are available on the UCSC genome browser UNIT 1 4 a link to these annotations will appear as well If the input sequence is taken from a genome assembly following this link is recommended If the target species is not listed select the organism that is closest Try several targets in different runs to see which one works best 6 Run N SCAN After specifying the input sequences the masking options and the target organism click the Predict Genes button in the lower dark green box under the sequence box to begin processing The browser now displays a Submission page that contains information on your input Fig 4 8 2 The submission has an I D number listed at the top of the page Below the target species the current state of your job is shown which can be Queued Masking Aligning Predicting or Complete The current status of the job is highlighted and an explanation of what it means is shown under Status When N SCAN is finished this page displays your results and a link is emailed to you At the top right side of the Submiss
65. not define the edges splice acceptor and splice donor sites of the exons To address these problems GRAIL 2 was developed Xu et al 1994a Instead of examining the coding potential of a segment of sequence of fixed length GRAIL 2 considers all segments of sequence between minimal splice junctions YAG for acceptors and GT for donors Though this initially presents a large number of candidates to the system that num ber is quickly reduced by applying a series of rules e g the requirement for at least one open reading frame per exon Those candidates that remain are screened by an artificial neural network that has been trained using a set of eleven parameters related to exon recognition this further reduces the set of potential candi dates After the candidates are assigned to clus ters the best candidate for each coding exon is selected GrailEXP uses the improved GRAIL 2 exon prediction system detecting 91 of all coding exons tegardless of size with a false posi tive rate of 9 In addition the system cor rectly predicts both edges for 61 of predicted exons and at least one correct edge for 96 of predicted exons Clearly locating coding regions in DNA sequence is critical to the interpretation of genomic DNA sequence Besides GRAIL a number of other approaches have been applied to the problem geneid Guigo et al 1992 UNIT 4 3 is a system that combines various rules and statistical features in an a
66. noteworthy feature of GeneMachine is that the process is fully automated the user is only required to launch GeneMachine and then open the resulting file with NCBI Sequin GeneMachine also does not require users to install local copies of the prediction programs enabling users to pass off to Web interfaces instead and reducing the overhead of maintaining the program albeit with the tradeoff of slower performance Annotations can be made to GeneMachine results prior to submission to GenBank thereby increasing the intrinsic value of the data A sample of the output obtained using GeneMachine is shown in Figure 4 1 3 and more details on this tool can be found on the NHGRI Web site http genemachine nhgri nih gov A recent paper by Makalowska et al 2002 illustrated the feasibility of identifying novel genes from regions of interest on chromosome 1 using GeneMachine as well in refining gene models and identifying interesting splice variants Finding Genes 4 1 7 Current Protocols in Bioinformatics Supplement 6 An Overview of Gene Identification 4 1 8 Supplement 6 FUTURE DIRECTIONS One of the most important questions arising from the completion of human sequencing is intimately related to the issues discussed in this overview What is the identity and precise location of all of the functional elements found within the human genome Determining the location of promoters transcriptional regulatory regions and factors
67. on which N SCAN has been installed and tested see Support Protocol with at least with a 2 GB memory and a processor whose computing speed is at least equivalent to a 2 GHz x86 processor Free disk space should be at least five times the combined size of the uncompressed target and informant sequences For example 2 GB is recommended for Arabidopsis using the current Brassica database as informant while 30 GB is recommended for a pair of assembled mammalian genomes Software An N SCAN software distribution See Support Protocol for obtaining and installing N SCAN Perl v 5 8 5 or later http www perl com Perl is already available on most Linux systems To check whether Perl is available and if so which version enter the following command at the Unix shell prompt perl v If Perl is not available those without substantial Unix experience should consult their system administrator RepeatMasker v 10 6 2006 or later is recommend also see UNIT 4 10 obtain RepeatMasker from hAttp www repeatmasker org RMDownload html N SCAN can be used without repeat masking but speed accuracy and disk usage will be affected Blastz v 12 27 2004 or later and multiz 4 28 2005 or later which can be downloaded from http www bx psu edu miller_lab The multiz package Current Protocols in Bioinformatics includes the lav2maf program which you will need for converting the blastz output Files The target sequence to be annotated an
68. one organism the directories should be named in a way that the user can keep track of the files e g MZEF_ HUMAN in the case of the human data set Current Protocols in Bioinformatics Finding Genes 4 2 7 Internal coding exons predicted by MZEF File Name m12523 fas Sequence length 19002 G C_ content 0 350 P Fri Fr2 Er3s OXE 3ss Cds 58s 0 992 0 528 0 460 0 656 111 0 561 0 557 0 689 0 635 0 636 0 522 0 374 112 0 443 0 557 0 567 0 993 0 405 0 322 0 567 221 0 536 0 505 0 587 0 971 0 391 0 596 0 513 212 0 538 0 573 0 646 0 932 0 363 0 553 0 522 211 0 547 0 518 0 545 0 999 0 642 0 592 0 481 112 0 536 0 622 0 607 0 999 0 636 0 486 0 463 122 0 574 0 581 0 553 0 997 0 626 0 493 0 403 122 0 541 0 558 0 597 0 999 0 453 0 617 0 544 212 0 576 0 588 0 548 0 757 0 414 0 498 0 556 221 0 460 0 537 0 719 1 000 0 587 0 482 0 399 122 0 548 0 532 0 719 0 864 0 502 0 591 0 509 212 0 526 0 574 0 457 0 678 0 501 0 463 0 594 221 0 462 0 570 0 562 0 999 0 667 0 451 0 515 122 0 483 0 613 0 609 0 514 0 489 0 434 0 544 221 0 539 0 540 0 493 Coordinates 1817 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13341 13425 13702 13799 14977 15115 15534 15757 16941 17073 17812 17874 Figure 4 2 3 Prediction results from the interactive Unix version of MZEF prior probability 0 04 overlap 0 Using MZEF to Find Internal Coding Exons 4 2 8 2
69. options type in the program name RepeatMasker on the command line For this example mta57 grouse RepeatMasker_file S RepeatMasker RepeatMasker The following contents will be returned SYNOPSIS RepeatMasker options lt seqfiles s in fasta format gt default settings are for masking all type of repeats in a primate sequence Choose from a number of options q Quick search 5 10 less sensitive 2 5 times faster than default nolow Do not mask low_complexity DNA or simple repeats div number Mask only those repeats lt x percent diverged from consensus seq species lt query species gt Specify the species or clade of the input sequence choose only one contamination options running options output options To get detailed help type in mta57 grouse RepeatMasker_file RepeatMasker RepeatMasker h Run RepeatMasker 7 Run command line version of RepeatMasker on the local system Q path to RepeatMasker el current dna fa For this example run mta57 grouse RepeatMasker_file RepeatMasker RepeatMasker species elegans current dna fa Because the example sequence is from C elegans the species elegans command is used so that the C elegans Repbase repetitive element library file is used The result files will be written into the directory RepeatMasker _file the same direc tory where the query sequence file s reside s For this example the result files include curren
70. prediction the sequence should be masked for interspersed repeats These repeats are degenerate copies of transposable elements and make up about a third of the human genome The coding portion of genes almost never contains interspersed repeats therefore masking results in better gene prediction By default a sequence is masked for interspersed repeat Low complexity and simple repeats are short repetitive sequences such as TATATATATA or GAGATAGAGAGA Genes sometimes contain such repeats so by default they are not masked Check the Mask Low Complexity regions checkbox to change this The user can mask out additional sequence by inputting the sequence to be masked in lowercase and checking the Mask Lowercase checkbox All boxes can be checked independently 5 Select the clade and organism of the target Under Clade is a drop down menu with a list of clades nematode fungus vertebrate in sect and plant Left click the mouse button on the Clade box to display the list of options then move the cursor to the appropriate option and left click the mouse button again Now select the species in the Species box in the same manner This list is constantly updated with newly sequenced genomes When the organism is selected N SCAN automatically selects the informant genome and displays it in the Informant box The current version of N SCAN automatically chooses mouse as the informant organism for human target sequences hu man as informant for a
71. scores of its exons geneid predicts only the coding fraction of a gene Thus geneid defines four classes of exons First Internal Terminal and Single corresponding to single exon or intronless genes A multiexon gene starts with First exon start codon to donor site followed by any number Current Protocols in Bioinformatics t date Wed Jan 17 18 01 10 2007 source version geneid v 1 2 geneid imim es Sequence examplei Length 32001 bps Optimal Gene Structure 1 genes Score 16 70 Gene 1 Forward 8 exons 470 aa Score 16 70 First 736 1130 6 14 02 4 86 1 91 17 70 0 00 AA 1 132 examplei_1 Internal 5504 5618 0 49 10 2 72 3 11 6 23 0 00 AA 132 170 examplei_1 Internal 8778 5961 1 13 00 1 43 0 86 13 14 0 00 AA 171 228 examplei_1 Internal 8730 8836 0 84 02 4 75 3 34 3 72 0 00 AA 229 264 examplei_1 Internal 13186 13256 0 46 11 1 21 5 38 5 00 0 00 AA 264 288 examplel_1 Internal 21287 21488 2 78 22 2 27 5 44 15 95 0 00 AA 288 355 examplei_1 Internal 29896 30019 1 56 10 1 27 0 56 14 90 0 00 AA 355 396 examplel_1 Terminal 31726 31947 3 30 00 1 78 0 00 19 32 0 00 AA 397 470 examplel_1 gt examplei_i geneid_v1 2_predicted_protein_1 470_AA MGTS GDHDDSF MKMLRSKMGKCCRHCFPCCRGSG TSNVG TSGDHENSFMKMLRS KMGKWC CHCF PCCRGSGKSNVGAWGDYDHS AFMEP RYHIRREDLDKLHRA AWWGK VPRKDLIVMLR DTDMNKRDKEK RTALHLAS ANGNSEVVQLLLDRRCQLNV LDNKKRTALI KAIQCQEDECV LMLLEHG ADRN IPDEYGNT ALHYA I YNEDKLMAK ALLLY GADIESKNKCGLTPLLLGVHE QKQQVVK FL
72. scoring exons geneid distinguishes four types of exons Initial ORFs defined by a start codon and a donor site Internal ORFs defined by an acceptor site and a donor site Terminal ORFs defined by an acceptor site and a stop codon Single ORFs defined by an start codon and a stop codon This corresponds to intronless genes geneid constructs all potential exons which are compatible with the predicted sites Coding potential geneid uses a Markov model of order five to compute the likeli hood of an exon sequence to be coding The model is estimated from both exon and in tron sequences The probability distribution of each nucleotide given the pentanucleotide preceding it is estimated in a set of known exon and intron sequences From the exon se quences this probability is estimated for each of the three possible frames and three transi tion probability matrices F F and F are computed F 515253545556 is the observed probability of finding hexamer 515253545556 with s in codon position j given that pentamer S 152535485 1S with s in codon position j An ini tial probability matrix F is estimated from the observed pentamer frequencies at each codon position From the intron sequences a single transition matrix is computed Fo as well as a single initial probability matrix Jo Then for each hexamer h and frame j a log likelihood ratio is computed Fi h LF h bee Fy A as well as for each pe
73. seia MT evidence gene_1 Cc AC004463 2 annotation mat a 4600 geneid evid ii AC004463 3 9200 13800 18400 23000 27600 32200 36800 41400 AC004471 1 Ht Aon 41400 46000 Figure 4 3 8 Improving gene prediction by using external information Basic Protocol 3 A Default geneid prediction on sequence example2 B geneid prediction when the exon coordinates of gene AC004463 3 are given to geneid C Ensembl annotation of the sequence Current Protocols in Bioinformatics Finding Genes 4 3 11 Supplement 18 example2 known_gene First 29058 29316 AC004463 3 example2 known_gene Internal 29425 29678 AC004463 3 example2 known_gene Terminal 30246 30350 AC004463 3 Since we are assuming that the exonic structure of the second gene is completely de termined all the exons in the GFF file must share the same group identifier The new prediction obtained by geneid appears in Figure 4 3 8 panel B This prediction is now very similar to the actual gene structure in this region of the human genome Using external information to investigate alternative splicing forms with geneid 3 Run geneid on the third example example3 fa geneid P param human3iso param samples example3 fa geneid predicts a six exon gene in the forward strand see Fig 4 3 9 panel A It is known that this gene has a number of splice isoforms Fagioli et al 1992 some of them being 9 31
74. sequences dna Prepare system 1 Download and install programs RepeatMasker Tandem Repeat Finder TRF cross_match and WU BLAST as well as Repbase library files RepeatMasker is a Perl script and can be put in any desired directory cross_match will be e mailed to users after contacting the authors With an account properly set up Repbase Update will assign a user name and password to download the repetitive library files For this example make a directory called repeat in the home directory and then copy RepeatMasker TRF and cross_match into this directory For this example type mta57 grouse mkdir repeat mta57 grouse cd repeat Current Protocols in Bioinformatics Finding Genes 4 10 7 Supplement 25 Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences 4 10 8 Supplement 25 2 Change the permission of the programs For this example type mta57 grouse repeat chmod u x RepeatMasker mta57 grouse repeat chmod u x crossmatch mta57 grouse repeat ln s trf321 linux exe trf 3 Set the correct paths by running the Configure Script First find out where Perl is installed mta57 grouse which perl usr bin perl Then after changing to the directory repeat and the directory RepeatMasker get the current directory path using the command pwd mta57 grouse RepeatMasker pwd home mta57 repeat RepeatMasker Then do the same for the TRF and cross match to
75. should be limited to those that are validated by laboratory evidence as opposed to computational predictions of genes These genes will Current Protocols in Bioinformatics form an adequate training data set if a sufficient number is found Unfortunately this is rarely the case for organisms targeted for whole genome sequencing therefore other methods should be used in order to construct a reliable data set In the authors experience an effective strategy for constructing a training set is to wait until a genome project has generated several hundred thousand base pairs of data From this data one can easily extract all of the long open reading frames ORFs i e stretches of DNA sequence without a stop codon Long ORFs may be 2500 bp depending on the GC content of the genome These long ORFs may then be searched against a non redundant protein sequence database using BLAST Altschul et al 1990 UNITS 3 3 amp 3 4 and any ORFs that have a significant hit may be safely assumed to be derived from real genes A step by step procedure to train GlimmerM follows Necessary Resources Hardware A Unix workstation GlimmerM has been successfully compiled for Linux Digital Unix and SunOS and it should be easy to compile on any platform supporting ANSI C and C Software GlimmerM 2 0 is an upgraded version that contains the automatic training procedure and a generally applicable gene finding algorithm GlimmerM 2 0 contains a
76. strand will have these coordinates listed in decreasing order Non coding exons and non coding portions of exons should not be listed for example if an exon spans positions 200 300 of a sequence and the start codon occurs at position 250 then the coordinate file should just list 250 300 A blank line must separate different genes The format of this file is iven in the example below g P Finding Genes 4 4 5 Current Protocols in Bioinformatics Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 6 Seql 1 15 Seql 20 34 Seql 50 48 Seql 45 36 Seq2 17 40 In this example Seq1 has two genes the first one is on the direct strand and its coding sequence covers positions 1 15 and 20 34 for a total of 30 nucleotides The second gene in Seq is on the complementary strand while Seq2 has only a single intronless gene on the forward strand GlimmerM can also use incomplete gene sequences in training provided that the coordinates given in this file start in frame For example suppose that the first gene on Seq in the above example extends off the sequence in the 5 direction with its unknown start codon somewhere upstream If the correct reading frame starts in position 2 then its exon coordinates should be specified in the training file as Seql 2 15 Seql 20 34 The FASTA file and coordinate file used in the example below are available at the Current Protocols Web site http www3 interscience wiley c
77. supporting Perl scripts The inputs to the N SCAN executable include a target sequence to be annotated a parameter file and an alignment between the target and informant sequence s Two supporting scripts Lav2maf and maf_to_align pl create the alignment file from a Blastz report Another script Nscan driver pl implements a simple four step pipeline for au tomatically masking the target sequence running Blastz running lav2maf and maf_to_align pl and running N SCAN Since Nscan driver p1 see Alternate Protocol 2 may require modification for a given user s needs and environment this pro tocol focuses on how to execute these four steps manually Once the manual procedure is understood users will be in a better position to understand whether Nscan_driver pl will work unmodified in their environment and if not what modifications are required Alternate Protocol 2 describes the use of Nscan_driver p1 but the authors strongly recommend reading Alternate Protocol 1 first Preparing Data Files and Running N SCAN Manually Starting with the data files described below there are five steps to preparing intermediate files and running N SCAN These five steps can also be orchestrated through a Perl script called Nscan_driver pl see Alternate Protocol 2 NOTE Users who are unfamiliar with Unix are encouraged to read APPENDIX 1C which provides guidance for working in a Unix environment Necessary Resources Hardware A computer
78. system gt lt FASTA file with the DNA sequence to be analyzed gt or glimmerm_ lt system gt lt FASTA file gt d lt directory of the training files gt where lt system gt is linux alpha or sun The d parameter specifies the directory containing the training files For the pre trained versions of the system this directory will be Glim merM trained_dir organism_name For user trained executables see Support Protocol this directory will be Train GlimmM date time Other optional parameters that can be given to the program are shown in Table 4 4 1 The annotations below discuss commonly used parameters The remaining parameters are discussed in the Critical Parameters section below The minimum gene length can be specified with the g option This value is the length of the smallest fragment considered to be a possible gene and is measured from the first base of the start codon to the last base before the stop codon The o and p parameters refer only to the special version of GlimmerM trained for malaria see the Files section and specify the amount by which two coding regions are allowed to overlap to be considered different gene models the default overlap length and percent are 30 bp and 10 respectively To determine if a putative model is likely to be a gene GlimmerM scores the coding region of that model in each of the six possible reading frames If the putative model s coding sequence in the correct
79. that contains two parameter sets target genome parameters and phylogenetic parameters The target genome parameters describe character istics of the genome to be analyzed such as the intron intergenic and UTR length distri butions the splice acceptor and donor sites and the hexamer composition of coding and noncoding sequence Since genomes from the same clade e g mammals usually have simi lar characteristics it is possible to use parame ter sets that were optimized for another species in the same clade However it is not advis able to use parameters that were optimized on a more distantly related species since their genome characteristics may be very different and this will have deleterious effects on ac curate gene prediction The phylogenetic pa rameters describe the patterns of divergence between two genomes Accuracy is less sensi tive to the phylogenetic parameters than to the target genome parameters If parameters are not available for a given genome pair parame ters from a pair with similar evolutionary dis tance and similar target genome gene density can be substituted The N SCAN Web site is regularly up dated as parameter sets are improved The Sub mission page for each N SCAN run lists the N SCAN code version and it also contains a link to the parameter file that was used This information is essential to reproducing results so one should be sure to include it in all com munications regarding N SCAN
80. that some are not An estimation of the gene finder s accuracy is presented below as well Malaria Version The output of the malaria specific version Fig 4 4 8 is somewhat different and will be explained first followed by an explanation that covers all other versions The first few lines of output from the Plasmodium malaria version of GlimmerM specify the settings of various parameters in the program Current Protocols in Bioinformatics ALTERNATE PROTOCOL Finding Genes 4 4 11 zt File Eda View Go Communicator e 3 2 wf S B Back Forward Reload Home Search Netscape Print Security Shop Stop g w Bookmarks Netsite http vvv tigr org tdb glinnern glar_forn htal EJT What s Related E ld ei ACO Docs Chapter 3 Functions AltaVista Search g Entrez Home g Logon Google GlimmerM Web Server In order to use the gene finder please select the organism for which you are doing the prediction then input your sequence by cut and pasting into the sequence window or enter a ename to upload Input sequences may be in FASTA format or simple DNA sequences Arabidopsis thaliana Organisms Oryza sativa rice Plasmodium falciparum malaria e Mail results No IT wait oYes to Sequence up to 200 000 bp predatod mpertea Malaria tenp fasta Brovee or paste the FASTA DNA sequence below up to 31 000 bp Please be patient it may take a while SS Sa Figure 4 4 6 Example
81. the a priori probabilities m and m_ for a randomly chosen sample being in class or respectively are known then the a posteriori probability q x of seeing the data x and it belonging to class is given by the Bayes formula q x T f 1 A Tf this is because g x p lx p x p x p xlt t p x pilt pal n p xl r_ A discriminant function h x is defined as the log likelihood ratio h x In q x One can choose the decision boundary C the Bayes decision rule as the hyper surface h x 0 because for any given sample point x it would be more likely to belonging to class if h x gt 0 By assigning x to class one would make an error with probability gx lt q x Similarly by assigning x to class when h x lt 0 one would make an error with probability g x lt q_ In general for any decision rule C the total error the Bayes error probability of misclassification J q_ x dx J q4 x dx R R 4 2 1 where the regions R and R_ are classified to and by C respectively QDA and its Relation to LDA When samples are assumed to be drawn from two different normal distributions 1 E f gt C Hy De ono 1 One Ae exp ae 4 2 2 Om 2y ee ge k where u and are the mean and the covariance matrix for the class k k or X is the determinant of the pxp matrix and A x y is called Mahalanobis distance between t
82. the organisms included in the 1 2 release might want to get both versions in order to compare the results The original GlimmerM system designed specifically for P falciparum uses a slightly different algorithm than subsequent versions of the GlimmerM program as explained in the Guidelines for Understanding Results and Commentary sections of this unit Because this initial algorithm had its own advantages the authors chose to keep it and include it in a separate directory as part of the software release After downloading the GlimmerM software one can find this initial gene finder including source code binaries and the latest malaria training set in a separate subdirectory called Malaria The source code for the current version of the gene finder can be found in the sources subdirectory and the training procedure is included in the train subdirectory Each subdirectory contains a Readme file explaining how to locally compile the source code Organism specific versions of the system can be found in the trained_dir subdirectory GlimmerM is available free of charge to researchers using it for non commercial purposes The system includes source code and a Readme file describing how to compile and train the system Pre trained versions for a small number of organisms Plasmodium falciparum Arabidopsis thaliana Oryza sativa Theileria parva and Aspergillus fumigatus are included that number continues to grow as more genomes are sequenced Inord
83. to identify the exons The example used in the following is a 19 kb human genomic DNA sequence containing the serum albumin ALB gene File name m12523 fasta GenBank accession no M12523 gi 178343 Minghetti et al 1986 The sequence may also be found on the Current Protocols in Bioinformatics Web site at http vww3 interscience wiley com c_p cpbi_sampledatafiles htm This gene has an alternative last exon the CDS annotation is as follows CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 18526 18555 Contributed by Michael Q Zhang Current Protocols in Bioinformatics 2003 4 2 1 4 2 18 Copyright 2003 by John Wiley amp Sons Inc UNIT 4 2 BASIC PROTOCOL 1 Finding Genes 4 2 1 T Search Results Microsoft Internet Explorer 10j x Fle Edt View Favortes Toos Help Stak gt O D D Aeh Gyrovotes Geto D Sw Sl address E hetp fargon lt sb orglog genefinder meet cop OG Unks E MZEF Results human gi 178343 gb N12523 1 HUMALBGC Human serum albumin ALB gene complete cds Tue Jan 22 17 45 06 2002 Strand 1 Overlap 0 Prior Prob 02 uploaded sequence file C mzhang ml2523 txt Internal coding exons predicted by MZEF Sequence_length 19002 G C_content 0 350 Coordinates P Frl Fr2 Fr3 Orf 3ss Cds 5Sss 1817 1854 0 98310 52810 4600
84. via E mail by typing in an E mail address before submitting The results are displayed on the browser in Figure 4 2 1 See Guidelines for Understanding Results below for analysis USING THE COMMAND LINE UNIX VERSION OF MZEF TO ANALYZE GENOMIC DNA SEQUENCES The software for the Unix command line version of MZEF can be downloaded from the anonymous FTP site ftp cshl edu pub science mzhanglab mzef This site contains a README file and three folders with human MZEF mouse mMZEF and Arabidopsis aMZEF versions of the program Necessary Resources Hardware Any Unix or Linux workstation Software The appropriate MZEF Command line executable file e g mzef_cmd_1mb_sun The executable files for MZEF are free for academic users The files may be downloaded from the cshl org FTP site see step 1 below Commercial users and those who wish to obtain source codes written in FORTRAN 77 should contact the CSHL licensing office Dr Carol Dempster 516 367 6885 dempster cshl org The software has evolved into many different versions to meet the demands from different users Consequently there are several executable files available from the FTP site The file names indicate the differences between the various forms The default platform is Sun Solaris unless indicated explicitly at the end of an executable file name The 1mb means the maximum input sequence size is 1 Mb otherwise the maximum is 200 Kb The cmd means all of the par
85. 0 00069719 0 4336 O0089650 00089654 0 9366 0 9949 00009507 00087706 1 gt ehr20 300001 400000 1 Predictions on the complementary strand No Promoter Pipromoter Exon Piexon Pidomor Cpe Window Pank 1 00090101 00069532 0 9500 00009601 00009264 0 9993 0 9909 00089713 00089S12 i 2 00077314 00076745 1 0000 00076014 00076752 1 0000 0 9814 00076917 00076716 1 3 00050372 00049803 1 0000 00049072 00049006 1 0000 0 9970 00050036 00049837 i 3 OOOSO26 00049697 1 0000 O004976 00049460 1 0000 0 9420 00050036 000496037 2 4 00016354 0001S785 1 0000 OO01S854 0001S278 1 0000 0 9593 00016621 00016420 i a Figure 4 7 2 Screen shot of FirstEF output of Example 1 with default cut off values for P exon P promoter and P donor For each sequence in the input file FirstEF presents predictions on direct and complementary strands separately Sequence header follows the symbol gt for each block of predictions The line immediately following gives the strand information for the predictions which follow Descriptions of each column are given in Table 4 7 1 Table 4 7 1 Description of the Columns in FirstEF Output Column Description No Serial number of the predicted first exon cluster Promoter Predicted promoter of length 570 bp P promoter A posteriori probability of promoter for a given window of size 570 bp Exon Predicted exon boundaries P exon A posteriori probability of exon for a given GT and promoter reg
86. 00 6200 san0 12400 15500 18600 21700 24800 27900 noc geneid i l EST1 geneid EST1 Isoforml Cc EST2 geneid EST2 Isoform2 D EST3 geneid EST3a geneid EST3b Isoform3 E Promoter geneid Prom o 3100 6200 00 12400 15500 18600 21700 24800 27900 IW Figure 4 3 9 Using external information to investigate alternative splicing forms with geneid Basic Proto col 3 A Default geneid prediction on sequence example3 B C Prediction of two alternative transcripts The EST1 and EST2 tracks display the exonic structure of partial ESTs matches whose coordinates have been given to geneid geneid EST1 and geneid EST2 show the resulting geneid predictions Isoform1 and Isoform2 correspond to the coordinates of the two isoforms D Prediction of a third alternative transcript The ESTS track displays the exonic structure of the EST whose genomic coordinates has been given to geneid geneid EST3a and geneid EST3b display the geneid predictions before and after the exon filtering process The Isoform3 track contains the annotation for this isoform E The coordinates of a promoter element Pro moter may be obtained by experimental means are given to geneid which improves the prediction of the first coding exon geneid Prom Using geneid to Identify Genes 4 3 12 Supplement 18 Current Protocols in Bioinformatics displayed in Figure 4 3 9 panels C D and E A
87. 2633 11583 10224 Refer to Table 4 7 2 for this and the exon coordinates of other mapped cDNA ESTs For the coordinates of the first exons and CpG windows refer to Figure 4 7 2 Necessary Resources Hardware Computer with Internet access e g PC running Microsoft Windows or Linux Apple Macintosh or Unix workstation Software Internet browser e g Netscape Navigator Microsoft Internet Explorer BLAST MEGABLAST http www ncbi nih gov BLAST UNITS 3 3 amp 3 4 SIM4 http pbil univ lyon1 fr sim4 html GENSCAN http genes mit edu GENSCAN html MZEF http www cshl edu mzhanglab UNIT 4 2 Files The DNA sequence of interest in FASTA format APPENDIX 1B 1 BLAST the sequence against the nr and EST databases using BLASTN UNIT 3 3 or MEGABLAST unr 3 4 in case of very long sequences Note the list of accession numbers from the BLAST output of cDNAs and ESTs with a percent identity score 299 2 Use SIM4 hittp pbil univ lyon1 fr sim4 html Florea et al 1998 to align each of the cDNA ESTs with the genomic sequence so as to identify exons with canonical splice sites Current Protocols in Bioinformatics Table 4 7 2 Annotations and Corresponding Exon Coordinates for Example 1 Annotation Mapped predicted exon boundaries AK027391 14015 12633 11583 10224 AJ335260 15142 15732 BI769142 15443 15615 18512 18533 21911 22419 AK026982 15814 15843 17950 18064 18339 18533 21911 23573 25046 25058
88. 3 0 6 493 34 6 3 4 1 6 hgl8_dna 3769 3921 18618 MIR SINE MIR 98 255 7 5 182 22 6 18 9 0 0 h hg 8_dna 4020 4072 18467 C MIR SINE MIR 122 140 7 7 342 27 0 9 0 4 5 hgi8_ dna 4349 4756 17781 LiMESE LINE Li 468 692 99 8 261 15 9 27 5 0 0 hgi8 dna 5500 5568 16971 C MIR SINE MIR 3 259 172 9 1373 15 0 0 9 2 6 hg i amp _dna 6279 6511 16028 MER30 DNA MER1_type 2 230 0 10 904 9 3 0 8 0 0 hgiS_dna 6635 6763 15776 LIPAL O LINE L1 6034 6163 5 11 400 30 5 9 4 1 7 hg i8_ dna 6884 7043 15496 MIR SINE MIR 79 250 18 12 327 32 5 2 5 0 8 hgi _ dna 7064 7184 15355 MIRb SINE MIR 140 262 6 13 383 34 2 4 6 4 1 hgl 8_dna 7260 7500 15039 C MIRc SINE MIR 8 260 19 14 282 22 8 7 4 5 8 hgi8_dna 9370 9504 13035 MIR SINE MIR 90 226 36 15 270 31 1 16 7 0 7 hgi8 dna 9611 9730 12809 C MIR SINE MIR 0 262 124 16 404 32 4 7 2 5 0 hgi8 dna 9798 9995 12544 MIRS SINE MIR 1 202 6 17 240 26 9 0 0 0 0 hgi8_dna 10016 10067 12472 GA rich Low_complexity 1 52 0 18 373 27 7 11 5 1 3 hgl8_dna 10123 10261 12278 C MIR SINE MIR 47 215 63 19 212 35 4 3 5 1 8 hgi8_ dna 10641 10780 11759 MIRc SINE MIR 101 238 24 20 571 29 8 7 3 2 5 hgi _dna 12043 12314 10225 C MER121 DNA TcMar 37 360 76 21 380 32 2 6 2 1 6 hgi8_dna 13353 13529 9010 C MIRb SINE MIR 58 210 26 22 2277 26 6 3 2 1 1 hgi8_dna 13549 14201 8338 LIMESA LINE Li 461 6127 46 23 7676 16 6 1 8 1 7 hgi8_dna 14243 16662 5877 C LIMC1 LINE Li 17 6316
89. 3 16969 17073 ooocoocoocorOCOCOCOCOCOCOCOCOCOCCOCOO oO Sequ 2 Frl 983 0 528 0 985 0 405 0 942 0 391 0 791 0 385 0 869 0 363 0 999 0 642 0 998 0 636 0 808 0 633 0 994 0 626 0 809 0 619 0 998 0 453 0 866 0 467 0 604 0 414 0 575 0 411 0 000 0 587 0 839 0 560 0 TOF Us502 0 508 0 501 0 998 0 667 0 955 0 680 0 ence_lengt Fr2 460 2322 596 608 3553 592 486 498 493 468 617 633 498 486 482 ATT x591 463 451 465 O oo Oo Oo OC Oo OC 0 CO CO OO 0 0 0 0 oO oO Fr3 656 567 z513 510 522 481 463 455 403 396 544 545 556 564 399 432 509 594 1515 520 Internal coding exons predicted by MZEF File Name m12523 fas h Orf TLT 221 212 212 211 112 122 122 122 122 212 212 221 221 122 122 212 221 122 122 19002 O Oo 0 0 2 eo 8 0 0 Oo OO Oo 8 2 3ss 061 536 2938 459 547 536 574 468 541 540 z916 548 526 G C_content D D GC O QOQ SS OO SS O GOG O OS O Cds 957 z509 2013 081 518 622 s9gl 983 558 2554 588 602 2937 2547 6932 922 574 s970 613 622 oo Qq Oo 2 co Co 0 Oo CO Co oO Oo oO 0 0 0 oO oS 5ss 689 587 646 646 545 607 993 093 997 597 548 548 pee TEY e PLS T9 457 562 609 609 0 350 Figure 4 2 2 Prediction results form the Command Line Unix version of MZEF prior probability 0 02
90. 3893 24 716 13 6 2 1 6 6 hgi8 dna 16665 16806 S733 C LiMCi LINE Li 2196 3950 3815 24 S501 30 4 2 6 5 8 hgl _dna 17885 18153 4386 MER112 DNA MER1_type 1 261 0 25 273 35 8 4 3 4 2 hgi8_dna 18242 18545 3994 C L2b LINE L2 31 3395 3087 26 567 24 7 6 5 0 6 hgl8_dna 19391 19545 2994 C LiMEd LINE Li 954 165 2 27 9766 1 7 0 1 0 0 hgi8g_ dna 19885 21017 1522 C LiPi LINE L1 1193 4953 3820 28 6237 2 3 0 0 0 0 hgi8_dna 21018 21744 795 LIPA2 LINE Li 5427 6153 2 29 415 20 5 1 2 0 0 hgl8_dna 22214 22296 243 C MERSSA DNA MER type 0 224 141 30 1020 23 5 1 4 0 5 h gi8 amp _dna 22316 22537 2 MER44C DNA MER2_type 9 232 01 31 hgi _dna 3 216 2 4 638 Dgi8_dna 490 705 21834 MIRDb SINE MIR Oto 2 355 32 713 0 0 8 gl _dna 1375 2464 20075 LIMC4a LIMNE Li 6740 7882 0 3 l 2773 21 0 6 0 1 2 hgi _d na 2598 2832 19707 HIRb SINE HIR 20 252 16 4 589 37 2 0 4 2 3 hgi _dna 3643 3726 16813 MIR SINE MIR 15 97 165 s 493 34 6 3 4 1 6 bgl8_dna 3727 3769 18771 TA n Simple_repeat 2 3 Ww 6 38 0 0 0 0 0 0 hgi _ dna 3769 3921 16618 MIR SINE MIR 98 255 m s 493 34 6 3 4 1 6 Dgi _dna 4020 4072 18467 C HIR SIWE MIR 122 140 7 7 182 22 6 18 9 0 0 hgif_dna 4349 4756 17761 LIMESE LINE Li 4680892 99 8 32 27 0 9 0 4 5 bgi _dn 5500 5568 16971 C MIR SINE MIR 3 259 an 9 261 15 927 5 0 0 hgi8_ana 6279 6511 16028 MERSO type 2 230 to 10 l 1373 15 0 0 9 2
91. 7 CDS 6591 6811 cbs 7815 8118 CDS UTR 15888 16218 MAMQREAGVQDFVLLDQOVSMEKFMDNLRKRFONGSIYTYIGEVCVSMNPY RQMNIYGPETIRKYKGRELFENAPHLFAIADSAYRVLKQRQQDTCILISG ESGAGKTEASKIIMKY IAAVTNAQGQNEIERVKNVLIQSNAILETFGNAK TNRNDNSSRFGKYMDIEFDYKADPVGGI ITNYLLEKSRVVQQQPGERNFH SFYQLLRGANDNELRQYELQKETGKYHYLNQGSMDILTEKSDYKGTCNAP KTLGFSTDEVQTIWRTIAAVLHLGNVEFQTIEDELVISNKQHLKSTAKLL QVTETELSTALTKRVIAAGGNVMOKDHNATQAEYGKDALAKAIYDRLFTW IISRINRAILFRGSKTQARFNSVIGVLDIYGFEIFDSNSFEQFCINYCNE KLQQLFIELVLKQEQEEYQREGIEWINIEYFNNKIICDLVEQPHKGIIAI MDEACLSVGKVTDDTLLGAMDKNLSKHPHYTSROQLKPTDKELKHREDFRI THYAGDVIYNINGF IEKNKDTLYQDFKRLLENSKDANLSEMWPEGAQDIK KTTKRPLTAGTLFQRSMADLVVTLLKKEPFYVRCIKPNDLKSSTVFDEER VEHQVRYLGLLENLRVRRAGFVHRQRYDKFLLRYKMISQYTWPNFRAGSD RDGVRVLIEEKKFAQDVKYGHTKIFIRSPRTLFALEHQRNEMIPHIVTLL QKRVRGWIVRRNFKKMKAAITIVRAYKAYKLRSYVQELANRLRKAKQMRD YGKSIQWPQPPLAGRKVEAKLHRMFDFWRANMILHKYPRSEWPQLRLQII AATALAGRRPYWGQARRWVGDYLANSQENSGYEAYNGS IKNIRNHPADGE TFQQVLFSSFVKKFNHFNKQANRAFIVSDSTIHKLDGIKNKFKDMKRTIK IRELTSISVSPGRDQLIVFHSSKNKDLVFSLESEYTPLKEDRIGEVVGIV CKKYHDLTGTELRVNVTTNISCRLDGKARI ITVEAASNVEVPNFRPKEGN IIFEVPAAYCV Show Transcript 2 3 4 5 6 7 8 9 Figure 4 8 4 An example of the bottom portion of the Submission Results Web page 9 Look at textual details and sequences for predicted genes Under the gene overview table details on every gene are listed Fig 4 8 4 A table shows the type orientation positive strand or negative strand start position en
92. 947 3 30 3 examplel_1l Graphical representation of the predictions Use the option save as over each individual picture i semati Postscript image Pictures developed with the program gff2ps by J F Abril and Roderic Guigo Figure 4 3 13 geneid Web server output with the sequence example1 fa The output is divided into two main areas the plain text output see Basic Protocol 1 step 6 and the graphical output of the predictions see Basic Protocol 2 step la Images are provided in JPG format although a PostScript document can be generated on the fiy by switching the Postscript Image button on At the bottom of the output information about the process parameters used and options is displayed HOW TO GET geneid AND VISUALIZATION PROGRAMS The geneid Web page is at http genome imim es software geneid index html From this page users can download the software and the accompanying documentation and obtain other information about geneid The geneid software can also be downloaded by anony mous FTP from ftp genome imim es The program geneid is written in ANSI C and runs on Unix based operating systems such as Linux MacOSX Solaris and Irix geneid source code compiled binaries parameter files and documentation are available under the GNU GENERAL PUBLIC LICENSE This protocol describes how to download and install geneid Necessary Resources Hardware Unix Linux workstation with at least 256 Mb RAM recom
93. 98a a 60 bp flanking intron window is used instead of 54 bp Since no isochore is found in Arabidopsis genome no G C specific feature variables are necessary Because of the G C content feature itself had been recognized as the important variable Arabidopsis_MZEF introduced one additional feature variable GC_ratio score x G C content in the exon G C content in the flanking introns Current Protocols in Bioinformatics Using geneid to Identify Genes The gene prediction program geneid is based on a simple hierarchical design 1 search splicing signals start codons and stop codons 2 build and score candidate exons and 3 assemble genes from the exons Guig6 et al 1992 Parra et al 2000 Geneid was one of the first computational gene identification programs an early version of geneid was available as an e mail server in 1991 Guig6 et al 1992 A new implementation of the system was written and released in 1999 geneid v1 0 Parra et al 2000 In 2002 new capabilities were added to geneid in order to include external information that supports genomic reannotation procedures geneid v1 1 A new more powerful version of geneid geneid v1 2 was released in 2004 including new parameter configurations for a larger number of species This version while having an accuracy comparable to the most accurate gene prediction programs is very efficient at handling large genomic sequences in terms of memory and speed geneid i
94. AATGAAATGAAAGGTCIGA AGATCAGAATGCAAAGCTGATCTGNNNNNNNNNNNNNNNNNNNNNNNNNN Figure 4 10 3 Web RepeatMasker result from an example run showing the Masked Sequence annotations section which lists the repetitive elements masked sequences replaced with Ns See Guidelines for Under standing Results for explanation Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences 4 10 4 Supplement 25 Select a method for returning results from the two radio buttons next to return method htm or email If html is selected in this step and html format was selected in step 2 above all of the results will be displayed in the browser If html is selected in this step and tar file was selected in step 2 the results will be provided as links in the browser If email is selected one should enter one s e mail address so that the results can be sent via e mail For this example select html At this stage one can choose to click the Submit Sequence button to start running RepeatMasker with the other options set at default values If the default settings do not satisfy one s needs continue with steps 5 to 8 and submit the sequence at step 9 For this example click Submit Sequence with other options set at default values The results that will be displayed on the browser are shown in Figures 4 10 2 4 10 3 4 10 4 and 4 10 5 See Guidelines for Understanding Resu
95. AGCAAGATGGGC date Wed Jan 17 18 15 02 2007 source version geneid v 1 2 geneid imin es Sequence example1 Length 32001 bps Firsts predicted in sequence example1 0 32000 First 186 174 6 10 01 0 59 1 40 1 54 0 00 7 MTAWLTt First 156 181 5 74 02 0 59 0 54 1 9 0 00 9 MTAWLTCNgg First 156 197 11 66 00 0 69 6 71 7 21 0 00 14 MTAWLTCNGCASLC First 178 181 6 53 01 2 09 0 54 0 00 0 00 2 Mg First 178 197 10 49 02 2 09 6 71 0 53 0 00 7 MGALRFgc First 178 216 11 93 00 2 09 6 71 4 12 0 00 13 MGALRFAFLAWPC Figure 4 3 2 Predicted Start codons top and First exons bottom on sequence example1 partial output The fields from left to right are defined in Table 4 3 1 and steps 3 and 4 of Basic Protocol 1 In the example the results of which are shown in Figure 4 3 2 top potential start codons are displayed by using the option b Other signals such as Stop codons Acceptor splice sites or Donor splice sites can be printed using the options e a or d respectively All options can be specified at once e g geneid bead in any order and geneid then produces the exhaustive list of all potential sequence signals For large and not so large genomic sequences this can produce very large outputs Each signal is printed in a separate record line with the following fields type of signal position score strand and signal sequence As geneid internally splits the input sequence int
96. AL519743 16138 16213 17950 18064 18339 18533 21911 22314 BG249690 Complement 16923 16799 BM042452 17001 17093 17950 18064 18339 18533 21911 22135 BM786491 17007 17093 17950 18064 18339 18533 21911 22065 AL136915 17948 18064 18339 18533 21911 22417 22432 23573 28702 28718 AL519782 17951 18064 18339 18533 21911 22530 BC001963 17953 18064 18339 18533 21911 23565 28702 28718 GENSCAN2 18346 18531 21909 22334 AL561818 22653 23474 AF116644 Complement 25060 25044 23565 23352 GENSCAN3 Complement 39001 38950 37758 37662 37426 37330 34439 34339 26114 26009 AK026945 49544 49585 49619 49870 56712 57002 59988 60280 64899 66260 67239 67251 BC019363 49764 49870 56712 57002 59988 60280 64899 66260 70389 70417 71182 71216 B1I253633 50014 50092 56712 57002 59988 60280 64899 65000 BQ439025 50059 50201 56712 57002 59988 60281 AF250311 56712 57002 59988 60286 64899 65391 GENSCAN4 50358 50360 56713 57003 59989 60281 64900 65392 The graphical depiction is presented in Figure 4 7 3 3 Submit the sequence to GENSCAN http genes mit edu GENSCAN html and or MZEF http www cshl edu mzhanglab UNIT 4 2 gene prediction programs and select the consensus predictions exons 4 Merge the exons belonging to overlapping cDNAs ESTs into a single transcript If there are no overlapping cDNA ESTs to support GENSCAN MZEF predictions co
97. AST tar xvf wu_blast directory will be seen after unpacking Programs within the wa_blast directory like blastp and blastx are executable after unpacking 2 Change the permission of the programs and the directories For this example mta57 grouse repeat chmod u x RepeatMasker mta57 grouse repeat chmod u x wu blast 3 Set the correct paths by running the Configure Script as described in Basic Protocol 2 To add a WU BLAST search engine enter Enter path home mta57 repeat wu blast 4 Create a new directory for input and output files RepeatMasker output files will be written to the same directory as the input file resides For this example type the following mta57 grouse repeat mkdir RepeatMasker_file mta57 grouse repeat cd RepeatMasker_file mta57 grouse RepeatMasker_file Next download or copy the FASTA file current dna fa gz for C elegans genome to the directory and unpack it mta57 grouse RepeatMasker_file gunzip current dna fa gz 5 Run program on command line using the flag w ublast For this example run mta57 grouse RepeatMasker_file S RepeatMasker RepeatMasker w species elegans current dna fa Here the flag w is used to indicate that WU BLAST is used as the matching engine the species elegans is used to indicate that the C elegans Repbase repetitive element library file is used since the sequence is from C elegans Note that species names that con tain multiple
98. An Overview of Gene Identification Approaches Strategies and Considerations Modern biology has officially ushered in a new era with the completion of the sequencing of the human genome in April 2003 While often erroneously called the post genome era this milestone truly marks the beginning of the genome era a time in which the availability of sequence data for many genomes will have a significant effect on how science is performed in the 21st century While complete human sequence data is now available at an overall accuracy of 99 99 the mere availability of all of these As Cs Ts and Gs still does not answer some of the basic questions regarding the human genome how many genes actually comprise the genome how many of these genes code for multiple gene products and where those genes actually lie along the complement of human chromosomes Current estimates based on preliminary analyses of the draft sequence place the number of human genes at 30 000 International Human Genome Sequencing Consortium 2001 This number is in stark contrast to previously suggested estimates which had ranged as high as 140 000 A number that is in the 30 000 range brings into question the one gene one protein hypothesis underscoring the importance of processes such as alternative splicing in the generation of multiple gene products from a single gene Finding all of the genes and the positions of those genes within the human genome sequence
99. Analysis Current Protocols in Bioinformatics Critical Parameters and Troubleshooting geneid is very easy to install and use and although is not bug free it should in general run without major problems In some cases however geneid behavior may not be what the user is expecting Mostly in these cases geneid will predict valid gene structures but users will be unhappy with them Unfortu nately it could also be that geneid does not produce results at all or that it crashes while running This section analyzes the most com mon causes of unsatisfactory geneid behavior and points to solutions whenever possible geneid runs correctly and produces a valid gene prediction but the user strongly suspect that the prediction is incorrect For sequences other than short ones encod ing single genes only in a few percent of the cases will geneid prediction be completely cor rect In most cases the geneid prediction will nearly reproduce at least one of the exonic structures of the genes encoded in the input DNA sequence A number of actual exons may be missed maybe more than when using other gene prediction programs and some false ex ons or genes may additionally be predicted in comparison to other gene prediction pro grams likely less In some cases the predic tion will certainly be disastrous There are a number of things the user can do to modify the default gene predictions If the coordinates of some of the coding e
100. Channel is beyond the scope of this unit users are referred to the above Web site Command line GrailEXP is available for download http compbio ornl gov grailexp gxpfaqg html by academic and nonprofit insti tutions allowing them to use the command line version which provides the most powerful ac cess to all the options and commands associated with GrailEXP A detailed description of the possible options and usages for the command line GrailEXP may be found in the GrailEXP Frequently Asked Questions FAQ Web page mentioned above The requirements for the command line version of GrailEXP include a Unix Linux Alpha Solaris Silicon Graphics workstation with at least 128 Mb RAM run ning Perl 5 0 or higher GrailEXP analysis in volves BLAST searches against several data bases This necessitates downloading and for matting the desired databases for use with GrailEXP Literature Cited Brunak S Englebrecht J and Knudsen S 1990 Neural network detects errors in the assignment of mRNA splice sites Nucl Acids Res 18 4797 4801 Brunak S Englebrecht J and Knudsen S 1992 Prediction of human mRNA donor and acceptor sites from the DNA sequence J Mol Biol 220 49 65 Claverie J M Sauvaget I and Bougueleret L 1990 K tuple frequency analysis From in tron exon discrimination to T cell epitope map ping Methods Enzymol 183 237 252 Dong S and Searles D B 1994 Gene structure prediction
101. Drosophila annotation experiment Genome Res 10 547 548 Brent M R and Guig R 2004 Recent advances in gene structure prediction Curr Opin Struct Biol 14 264 272 Castellano S Novoselov S V Kryukov G V Lescure A Blanco E Krol A Gladyshev V N and Guig R 2004 Reconsidering the evolution of eukaryotic selenoproteins A novel nonmammalian family with scattered phyloge netic distribution EMBO Rep 5 71 77 Castellano S Morozova N Morey M Berry M J Serras F Corominas M and Guig R 2001 In silico identification of novel selenopro teins in the Drosophila melanogaster genome EMBO Reports 2 697 702 Fagioli M Alcalay M Pandolfi P P Venturini L Mencarelli A Simeone A Acampora D Grignani F and Pelicci P G 1992 Alterna tive splicing of PML transcripts predicts coex pression of several carboxy terminally different protein isoforms Oncogene 7 1083 1091 Glockner G Eichinger L Szafranski K Pachebat J A Bankier A T Dear P H Lehmann R Baumgart C Parra G Abril J F Guig R Kumpf K Tunggal B Cox E Quail M A Platzer M Rosenthal A Noegel A A Dictyostelium Genome Sequencing Con sortium 2002 Sequence and analysis of chro mosome 2 of Dictyostelium discoideum Nature 418 79 85 Current Protocols in Bioinformatics Guig R 1998 Assembling genes from predicted exons in linear time with dynamic programmin
102. Figure 4 5 2 The text output from the GeneMark program for the example sequence Open reading frames predicted as genes are listed for both strands along with the average probability for an ORF and start probability for a translation start upstream to the putative start to be an RBS site For each possible start the program chooses the hexamer with the best score at a distance 4 to 21 nt from a putative start The position of the rightmost direct strand or leftmost complement strand nucleotide of the hexamer with the best score and its sequence are shown If the start site is adjacent to the edge of the sequence it is not possible to evaluate the RBS site and no data are given The minimum size of an ORF to be reported is half of the chosen window size parameter c List of Regions of Interest GeneMark can identify so called regions of interest i e areas between in frame stop Prokaryotic Gene codons with a high coding potential The format of this list is similar to that for the open Prediction Using reading frames list sample not shown GeneMark and GeneMark hmm 4 5 4 Supplement 1 Current Protocols in Bioinformatics gt Escherichia coli K12 Region 1 50000 Order 4 Window 95 Step 12 6 25 10 osF 4445445 oes e e e le b OSre 4 TEER In uM n se eS Sy a aa Comte 5 y Direct Sequence OSSh _ Irin Irmi m a I ma we Complementary Sequence 0 0 8800 9200 9800 10000 10400 Nucleoti
103. Genes in Human Genomic DNA n Stanford Univeristy Stanford University Stanford Calif Burge C and Karlin S 1997 Prediction of com plete gene structures in human genomic DNA J Mol Biol 268 78 94 Finding Genes 4 8 15 Supplement 20 Using N SCAN or TWINSCAN 4 8 16 Supplement 20 Elsik C G Mackey A J Reese J T Milshina N V Roos D S and Weinstock G M 2007 Creating a honey bee consensus gene set Genome Biol 8 R13 Flicek P Keibler E Hu P Korf I and Brent M R 2003 Leveraging the mouse genome for gene prediction in human From whole genome shotgun reads to a global synteny map Genome Res 13 46 54 Gross S S and Brent M R 2006 Using multiple alignments to improve gene prediction J Comput Biol 13 379 393 Guigo R Agarwal P Abril J F Burset M and Fickett J W 2000 An assessment of gene prediction accuracy in large DNA sequences Genome Res 10 1631 1642 Guigo R Dermitzakis E T Agarwal P Ponting C P Parra G Reymond A Abril J F Keibler E Lyle R Ucla C Antonarakis S E and Brent M R 2003 Comparison of mouse and human genomes followed by experimental verification yields an estimated 1 019 additional genes Proc Natl Acad Sci U S A 100 1140 1145 Guigo R Flicek P Abril J F Reymond A Lagarde J Denoeud F Antonarakis S Ashburner M Bajic V B Birney E Castelo R Eyras E Ucla
104. IKKKANLNVLDRYGR ICELLSDYKEKQMLK ISSENSNPVITILNIKLPLKV EEEI KKHGSNP VGLP ENLTNGASAGNGDDGLIPQRRSRK PENQQFPDTENEEYHSDEQND TRKQLSEEQNTGISQDEILTNKQK QIEVAEQKMNSELSLSHKKEEDLLRENSVLQEEIAM LRLELDETKHQNQLRENKI LEEIESVKEK TDKLLRAMQLNEEALTKTNI Figure 4 3 1 Default geneid prediction on sequence example The fields from left to right are defined in Table 4 3 1 possibly zero of Internal exons acceptor site to donor site and ends with a Terminal exon acceptor site to stop codon An intronless gene is constituted by a Single exon start codon to stop codon Lines starting with the character do not correspond to coding exons but provide additional information about the prediction At the top of the output two lines starting with the characters display general information on the geneid process After this main header the line beginning with Sequence displays the name and the length of the input sequence whereas the line starting with Optimal Gene Structure contains the number of genes predicted along the input sequence as well as the total score of the prediction which is the sum of the scores of the predicted genes Then lines starting with Gene provide general information on each gene gene identifier strand forward or reverse number of exons gene product length and gene score After this there is a line for each coding exon in the gene with the fields from left to right defined as in Table 4 3 1 After the set of line
105. Information Resources Projects amp Research Channel Generation Grail e Grail ExP e Pipeline Parser PROSPECT ORNL Genome Analysis Pipeline Results for Query D 1059421518 28565 Organisa a ang peer hauast ae ize Alabo bytes erat bp Retrieve All Rawoutput JavaViewer Coming soon Html Viewer 4 found Proteins r an 1 found Proteins Pian Breil EXP CpS Xelends _ thNAscan SE CNA Ganes Grail BAC Pairs Figure 4 9 3 Pipeline summary page The Java Viewer link allows the user to download a Java Swing applet for graphical display and visualization of the pipeline results The applet Fig 4 9 5 consists of three sections 1 Feature Display which displays genes and other identified biological features in a graphical form the horizontal gray scale bar in the center of the top Feature Display window represents the local GC content of the input DNA sequence with brightness proportional to the percentage GC content of that region of the sequence 2 Sequence Display which displays a 100 base double stranded region corresponding to the stretch of DNA selected using the scroll bar at the bottom of the Sequence Display and 3 Features Pane which is a tabbed pane with a table for each of the analysis features Each table displays the list of feature elements for that feature type and their locations scores and other relevant information Current Protocols in Bioinformatics d
106. J ENSEMBL such as mRNAs ESTs and other genomic annotations the user can combine that information with FirstEF predictions to produce more reliable first exon and promoter annotations for the human genome The procedure outlined in this protocol uses BLAST UNITS 3 3 amp 3 4 or MEGABLAST unr 3 3 SIM4 Florea et al 1998 and a gene prediction program either GENSCAN MZEF unir 4 2 or both to find probable internal exons and then combine them with FirstEF predictions Current Protocols in Bioinformatics ALTERNATE PROTOCOL SUPPORT PROTOCOL Finding Genes 4 7 5 Supplement 1 Application of FirstEF to Find Promoters and First Exons in the Human Genome 4 7 6 Supplement 1 res ee oeoo eere eee a ee lt Exons ren ATOM ASS E e oO o eS Ee BI 7est porn G w H 0 OHS bis pemo Cm 2 197 toi v t aed o BG 116644 W231 4 CE a ia fro re t BTs r p 1969 eo AS19782 n o Boos 96 i CO ra Genscan Genecard ARK ae ATALI MTK A nike oe 4 oN H _ Cph Figure 4 7 3 Part of the annotations for Example 1 from 1 to 60000 bp obtained by following the steps described see Support Protocol Exons and Exons indicate the predicted first exons on direct and complementary strands respectively The exon with symbol i j represents j ranked first exon in the i cluster AK027391 represents the transcript mapped to the exons complement 14015 1
107. KEEDLLRENSVLQEEI AM LRLELDETKHQN QLRENKILEEIESVKEKTDKLLRAMQLNEEALTKTNI Figure 4 3 3 geneid prediction in extended format Current Protocols in Bioinformatics Finding Genes 4 3 5 Supplement 18 BASIC PROTOCOL 2 Using geneid to Identify Genes 4 3 6 Supplement 18 gff version 2 date Wed Jan 17 18 33 02 2007 source version geneid v 1 2 geneid imin es Sequence examplel Length 32001 bps Optimal Gene Structure 1 genes Score 16 70 Gene 1 Forward 8 exons 470 aa Score 16 70 example1 geneid_v1 2 First 736 1130 6 14 0 example1_1 example 1 geneid_v1 2 Internal 5604 6618 0 49 1 example1_1 example1 geneid_v1 2 Internal 5778 5951 1 13 0 examplei_1 example1 geneid_vi 2 Internal 8730 8836 0 84 0 examplei_1 example1 geneid_vi 2 Internal 13186 13256 0 46 1 examplei 1 example1 geneid_v1 2 Internal 21287 8 21488 2 78 2 examplei_i examplei geneid_vi 2 Internal 29896 30019 1 56 1 examplei_1 example1 geneid_vi 2 Terminal 31726 31947 3 30 0 examplei_1 Figure 4 3 4 geneid prediction in GFF format file the GFF header Then following the same structure as geneid default format lines starting with the character assumed to be free format comments in GFF are used to provide general information on each gene predicted GFF records provide information about the predicted gene features from left to right sequence name source the gene prediction program geneid in
108. M pees mm pone A Reference Borodovsky M and Lukashin A unpublished UPDATE Ma 02 O sativa Rice Eukaryotic GeneMark hmm model has been updated Result of last submittal GeneMark beem Listing Go to Gent Mark hmm Protein Translations Go to Job Sebmittal GescMark ban Versio Exon t anger fiance 3 24 2 Figure 4 6 2 The text output for the Eukaryotic GeneMark hmm program Albumin CDS CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 18526 18555 Alloalbumin Venezia CDS CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 17688 17732 Figure 4 6 3 The DNA sequence of complete human serum albumin ALB gene GenBank accession no M12523 was used as sample input for the program The last exon of this gene shown in bold is alternatively spliced as either human serum albumin or human alloalbumin Venezia Finding Genes 4 6 3 Current Protocols in Bioinformatics Supplement 1 Eukaryotic Gene Prediction Using GeneMark hmm 4 6 4 Supplement 1 Large DNA sequences should be split into smaller ones that are more homogeneous in GC composition For example for H sapiens the recommended sequence size is 100 kb On the other hand very short sequences should be avoided To decreas
109. Obtain the predicted genes in the GFF format by using the option G geneid P param human3iso param G samples examplel fa gt geneid_output gff General Feature Format or GFF http www sanger ac uk Software formats GFF is a proposed standard format for describing genes and other features associated with DNA RNA and protein sequences Each feature is described as a list of fixed fields or columns delimited by tabs This format is very easy to parse by bioinformatics applications There are a number of tools including visualization ones that can process GFF files see Basic Protocol 2 geneid produces GFF compliant output with the option G This option can be applied to any set of gene features selected to be printed as in steps 2 and 3 The result is shown in Figure 4 3 4 A set of standardized lines appear at the top of the GFF date Wed Jan 17 18 28 20 2007 source version geneid v 1 2 geneid imim es Sequence example1 Length 32001 bps Optimal Gene Structure 1 genes Score 16 70 Gene 1 Forward 8 exons 470 aa Score 16 70 Start 736 738 4 86 GCGGCAAGAGCAACATGGGC First 736 1130 6 14 02 4 86 1 91 17 70 0 00 AA 1 132 examplei_i Donor 1130 1131 1 91 GAGGTAACC Acceptor 5503 5504 2 72 0 00 0 00 CCTCAAGTCTTCTCACTCTCATAGGAC Internal 5504 5618 0 49 10 2 72 3 11 6 23 0 00 AA 132 170 example1_1 Donor 5618 5619 3 11 AAGGTATGC Acceptor 5777 5778 1 43 0 00 0 00 TIGTTTTTGGTCTAATACTGACAGGCC Inte
110. Protocols in Bioinformatics Using N SCAN or TWINSCAN to Predict Gene Structures in Genomic DNA Sequences Marijke J van Baren Brian C Koebbe and Michael R Brent Washington University St Louis Missouri ABSTRACT N SCAN is a gene prediction system that combines the methods of ab initio predictors like GENSCAN with information derived from genome comparison It is the latest in the TWINSCAN series of programs This unit describes the use of N SCAN to identify gene structures in eukaryotic genomic sequences Protocols for using N SCAN through its Web interface and from the command line in a Linux environment are provided Detailed discussion about the appropriate parameter settings input sequence processing and choice of genome for comparison are included Curr Protoc Bioinform 20 4 8 1 4 8 16 2007 by John Wiley amp Sons Inc Keywords N SCAN e TWINSCAN e gene prediction e sequence alignment e comparative genome analysis e cross species sequence comparison e genome annotation INTRODUCTION N SCAN is a gene structure prediction system for eukaryotic genomic sequences N SCAN is the latest in the TWINSCAN series of programs It takes as input a sequence to be annotated the target sequence and a multiple sequence alignment of the target sequence and one or more closely related genomes the informant genomes N SCAN models the pattern of tolerated substitutions insertions or deletions in the aligned se quences an
111. RAM recommended Software geneid v1 2 full distribution see Support Protocol Unix text editor Files This protocol uses the two human genomic sequences listed below which were extracted from the UCSC human genome browser assembly March 2006 These sequences can be found at the samples subdirectory within the geneid distribution see Support Protocol and also at the Current Protocols in Bioinformatics Web site at http www currentprotocols com example2 fa 47 kb Location Human chromosome 22 coordinates 17 499 857 17 546 853 reverse strand example3 fa 32 kb masked Location Human chromosome 15 coordinates 72 071 133 72 117 117 example2 evidences gff example3 EST1 gff example3 EST2 gff example3 EST3 gff and Current Protocols in Bioinformatics example3 promoter gff contain annotated gene features on the above sequences 1 Run geneid on the second example example2 fa geneid P param human3iso param samples example2 fa geneid predicts a 21 exon gene on the forward strand Figure 4 3 8 panel A displays the default geneid prediction using gff2ps The region actually encodes three different genes all of them sharing exons with the geneid prediction For the example however assume that at the time of the prediction only one of these genes the second has been determined By providing the exonic structure of this gene the overall geneid prediction in this region improves substantially 2 Include
112. Splice site recognition When designing the splice site detection module of the gene finder the authors only considered the problem of recognizing the highly conserved dinucleotides GT and AG at the 5 and 3 intron boundaries While these dinucleotides are almost always present at the splice sites others appear with lower frequency Finding Genes 4 4 17 Table 4 4 3 GlimmerM s Performance On ARASET a Set of Genes from Arabidopsis thaliana Correct gene predictions Correct start sites Sn Sp nucleotide level Correct predicted exons true exons GlimmerM results 63 107 0 95 0 94 766 860 Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 18 in at least some eukaryotic genomes For ex ample the human genome contains a small number of AT AC introns which use a different set of splicing molecules The GT AG introns are spliced by a U2 type spliceosome while the AT AC introns are excised by a novel U12 type spliceosome Dietrich et al 1997 At least 11 distinct U12 type introns have been identified in Arabidopsis thaliana Wu and Krainer 1996 and many more are likely to be charac terized in the future Arabidopsis also contains a relatively high number of GC AG introns which use the normal splicing machinery but not the normal 5 dinucleotide The Arabidopsis data the authors used for training GlimmerM contained 2 AT AC introns and 48 introns with GC AG borders Despite the
113. T 76 00 5 00 67 00 00 68 00 87 00 59 00 76 00 86 00 71 00 80 00 59 00 Bad overlap vith gene 5 Figure 4 4 7 Output of GlimmerM Web Server predator nper tea hSof tuare Gl immerti Malariash i s ine x predator npertes Malari Minimum gene length 17 Hinimas 3601 predator npertea tiSof tuare Glinmert Malaria Figure 4 4 8 Sample output from the malaria specific version of GlimmerM The FASTA file used to generate this output is available on the Current Protocols Web site htto Avww3 interscience wiley com c_p cpbi_sampledata files htm 4 4 13 Current Protocols in Bioinformatics Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 14 Use independent scores indicates whether the module that scores each fragment using independent base probabilities is used in the interpolated Markov model scores see Background Information The parameters shown in Figure 4 4 8 have the default values but they could have been changed if the optional parameters described in Table 4 4 1 were used When GlimmerM is running then it will use decision trees see Background Information in computing the splice site scores only if these flags were created by the training procedure The next portion of the output in Figure 4 4 8 shows the scores of GlimmerM for each open reading frame in a given gene The first column represents an ID number for reference purposes IDs are assigned to all predicted gene
114. TCCATGC To see our predictions for human Click Here All selections made Gene prediction takes a while After submitting you will be redirected to a page where you can follow the progress and see the results If you would like your results emailed to you feel free to say so by setting your preferences If the results have been useful to you please cite Gross SS Brent MR Using multiple alignments to improve gene prediction In Proc 9th Intl Conf on Research in Computational Molecular Biology RECOMB 05 374 388 and J Comput Biol 2006 Mar 13 2 379 93 Korf Flicek P Duan D Brent MR Integrating genomic homology into gene structure prediction Bioinformatics 2001 17 Suppl 1 8140 8 Figure 4 8 1 The N SCAN Web server at hitp mblab wustl edu nscan lf you are not registered a Register link will be visible in the top right corner It is possible to do an example N SCAN run before registering Go to http mblab wustl edu nscan submit You will see a page very similar to Fig 4 8 1 but the Predict Genes button is replaced with See an example of Twinscan N SCAN in action Clicking on this link starts an example run which selects a Drosophila melanogaster sequence and runs N SCAN with default settings The sequence contains 6 genes which are all predicted correctly 2 Log onto the server at http mblab wustl edu nscan login Type in your username and password and press the Login button The N SCAN Web pag
115. TG Vertical ticks under the 0 5 level represent one of the three stop codons TAA TGA or TAG The thick gray horizontal line indicates a region of interest GeneMark generates the graphical output in Adobe PostScript format which is sent by E mail to the user Some E mail readers will allow the figure to be displayed automatically while others may require the installation of additional software programs To create a valid PostScript file one needs to select every line inclusive between PS Adobe 2 0 and EOF and save it to a file called graph ps or similar The file should not contain any blank lines before the PS Adobe 2 0 line or after the EOF line This file can then be viewed with any PostScript viewer program USING GeneMark hmm FOR PROKARYOTIC GENE PREDICTION The GeneMark hmm program Lukashin and Borodovsky 1998 Besemer et al 2001 uses a hidden Markov model framework with gene boundaries modeled as transitions between hidden states In contrast to GeneMark see Basic Protocol 1 the prediction of gene starts is further automated and human expert intervention to assess RBS sites and start codons is not required GeneMark hmm can be accessed through the Web Fig 4 5 4 It is convenient to use GeneMark hmm for sequences with multiple genes up to complete genomes where the speed of manual analysis becomes a bottleneck GeneMark hmm also uses precomputed models for 55 species as of August 2002 An
116. The default platform is SUN Solaris unless indicated explicitly at the end of an executable file name The 1mb means the maximum input sequence size is 1 Mb otherwise the maximum is 200 Kb The cmd means all of the parameters must be entered from the command line other files are interactive i e the program will prompt users for each parameter one line at a time during execution The static means it does not require a run time FORTRAN library the default requires 1ibF77 so x libraries The new or any versions after that 1997 or later will not require files and data being in the current directory to run Other versions may also be compiled at a special request to mzhang cshl org Files A FASTA file APPENDIX 1B with no more than 80 characters per line that contains the DNA sequence in which one wishes to identify the exons MZEF can only take the standard DNA RNA character symbols either in capital or lower case letters ambiguous IUPAC symbols APPENDIX 1A will be converted to the standard symbols by a random draw e g N will be converted into A C G T with equal probability Current Protocols in Bioinformatics The example used in the following is a 19 kb human genomic DNA sequence containing the serum albumin ALB gene File name m12523 fasta GenBank accession no M12523 gi 178343 Minghetti et al 1986 The sequence may also be found on the Current Protocols in Bioinformatics Web site at
117. a while genes predicted on the complementary strand are indicated with a minus sign If both Typical and Atypical models were selected genes predicted by the Typical model will be denoted as Class I and genes predicted by the Atypical model as Class 2 If only one type of model was selected only Class 1 will appear regardless of the model type If an incomplete gene was predicted a less than lt or greater than gt symbol will appear next to a coordinate for the left and right prediction edge respectively 6 Interpret the graphical output The GeneMark hmm graphical output is always combined with the GeneMark see Basic Protocol 1 graphical output Fig 4 5 6 The predictions made by GeneMark hmm are depicted by the thick black horizontal line on the axes of the six GeneMark panels If both the Typical and Atypical models were selected these horizontal lines will be shown in solid black and in dashed red fashions respectively If only one model was selected the predictions will be shown by a solid black line USING THE HEURISTIC APPROACH FOR PROKARYOTIC MODEL BUILDING The Heuristic algorithm Besemer and Borodovsky 1999 with the Web interface shown in Figure 4 5 7 builds an inhomogeneous Markov model of protein coding regions for an anonymous sequence based on the observed relationships between the positional nucleo tide frequencies and the global nucleotide frequencies observed in the analysis of 17 complete
118. a the R option see Basic Protocol 3 Users can modify the gene model to consider other features but the predicted features must be passed to geneid also via the R option Modification of the gene model may not involve the introduction of new features but changing the rules affecting default features for instance to force the prediction of only one gene COMMENTARY Background Information tributed to this first version of geneid This ver sion was developed at the Molecular Biology History Computer Research Resource Dana Farber Cancer Institute Harvard University It was never distributed but an E mail server was set up in late 1991 which was latter moved to the Biomolecular Engineering Research Cen ter Boston University Kathleen Klose and Steen Knudsen developed the server In 1995 a Web server was set up at the Institut Municipal d Investigaci M dica IMIM in Barcelona Mois s Burset developed the server Version 1 0 of geneid Parra et al 2000 was completely rewritten at the IMIM This version maintained the hierarchical structure signal to exon to gene in the original geneid but the scoring schema was simplified and furnished with a probabilistic meaning as discussed above A new version of geneid v1 1 was released in 2002 This version had a substantially improved engineering design which makes it more robust faster and more The program geneid Guig et al 1992 was one of the first programs to pre
119. akefile Makefile to build the binary file README Before starting to work with geneid it is necessary to compile the program i e produce a binary file properly generated according to the computer architecture For that move to the geneid directory by typing scd geneid To compile the program building a binary file type smake A new directory called bin has been created now Inside the geneid program is ready to be executed by the users Just to test the program showing the list of available options try the command S bin geneid h Current Protocols in Bioinformatics Finding Genes 4 3 19 Supplement 18 Using geneid to Identify Genes 4 3 20 Supplement 18 On most Unix systems this should be fairly simple but if you encounter problems please contact the authors at geneid imim es Throughout this unit for simplicity the relative path bin has been omitted in the examples just running geneid It is also advisable to set the GENEID environmental variable to point to the param subdirectory within the geneid directory The geneid distribution includes complete and exhaustive documentation It has been written in HTML and it can be accessed through a Web browser The documentation is also available at the geneid homepage How to get gff2ps The gff2ps Web page contains the information required to download and install the pro gram http genome imim es software gfftools GFF2PS html gff2ps can also
120. al discriminant surface non linear to separate them Zhang 1997 For a more detailed discussion of the theory behind MZEF please see this unit s Appendix Advantages MZEF is simple and fast It is easily portable and may be incorporated into other programs readily It can find internal coding exons in a short DNA sequence fragment that may not contain the full gene it only requires a 54 bp flanking intron sequence It can also output exons with alternative splice sites by allowing overlaps It can handle very short exons gt 18 bp and tends to give better accuracy on exon level statistics Limitations Since MZEF is only designed to identify one class albeit the most important class of exons internal coding exons one would need other tools for identifying the other eleven classes Zhang 1998c of exons see Suggestions for Further Analysis MZEF does not produce a gene model one has to assemble a gene model by hand This may not be regarded as a limitation when one is facing alternative splicing that occurs in nearly 60 of human genes IHGSC 2001 Modrek and Lee 2002 The user cannot adjust various threshold values other than the few input parameters and must run the reverse strand separately Other options for similar analysis There are two related programs that extend MZEF to improve performance under specific conditions One is called GSA2 X Q Huang unpub which has combined MZEF with the EST database search results
121. am R samples example3 EST2 gff samples example3 fa The genomic coordinates of the alignment of a different EST EST2 are given now to geneid The corresponding GFF file is example3 EST2 Internal 27330 27588 example3 EST2 Terminal 28652 28830 The prediction incorporates the two exons in the EST sequence Figure 4 3 9C and resembles closely another of the known alternative forms for this gene 6 Use geneid to obtain an alternative structure supported by a different EST geneid P param human3iso param R samples example3 EST3 gff samples example3 fa The genomic coordinates of the alignment of a different EST EST3 are given now to geneid The corresponding GFF file is example3 EST3 Internal 19031 19101 example3 EST3 Terminal 30180 30233 The resulting prediction appears in Figure 4 3 9D geneid EST3a As it is possible to see the geneid predictions include new exons between the two exons corresponding to the EST sequence The resulting prediction is thus incompatible with the EST sequence Grouping the EST sequences into a gene would certainly prevent the inclusion of these exons In the current version of geneid however grouped features cannot be extended Although the procedure is somehow more complex it will also serve to illustrate the option O which allows geneid to produce gene predictions from sets of exons provided externally Essentially the user must predict the exhaustive list of exons along the
122. ame the frames of GRAIL exons that match the alignments and spliceability Current Protocols in Bioinformatics among the various alignments An elaborate dynamic programming algorithm is then used to construct gene models Each node in the dynamic programming model isa GRAIL exon candidate or an alignment based exon Connec tions between the various nodes are scored with bonus for good connections and penal ties for bad connections Once the highest scoring dynamic programming model has been calculated the preliminary gene models are refined In this phase genes can be split or merged and the exon edges are tweaked to better match splice sites Alternative splices are then identified and added to the gene table In the next phase potential coding starts and stops within each mRNA are identified and evaluated to determine the correct one In the absence of protein similarity evidence this is not a perfect process In addition false stop codons in the genomic sequence because of sequencing er rors or pseudogenes can result in incorrect identification of start and stop codons Finally poly A sites and promoters are lo cated Poly A recognition uses a simple Markov model The promoter system uses a neural net and looks for specific types of signals TATA CAAT GC box Not every gene model will have a poly A site or a promoter The poly A site recognizer scans the sequence for the pattern AATAAA It then
123. ameters The critical parameters are the cut off values of a posteriori probabilities of promoter exon and donor site The closer the values of these parameters to the more likely the correspond ing predictions are real Troubleshooting The following is a list of some of the more common problems associated with using this software 1 Too many predictions Increase the cut off values or consider only those predictions with probability values close to one 2 Too few or no predictions This may be due to a lack of first exons in the genomic sequence that is being analyzed However if the user has strong reason to believe that there exists a first exon or promoter of some gene of interest experimenting with lower probability cut off values may produce some predictions 3 Promoter of single exon genes Single exon genes lack splice site GT and hence Fir stEF may fail to predict such promoters In such cases lowering cut off value of P donor would help to predict the promoter region 4 First exons with noncanonical splice sites FirstEF was trained on first exons with canonical splice site GT Hence FirstEF can not predict first exons with noncanonical splice sites Lowering cut off value of P donor might predict a nearby weak donor site and help identifying the promoter region Suggestions for Further Analysis Among the earlier programs PromoterIn spector Scherf et al 2000 is the best in locat ing the gene regula
124. ameters must be entered from the command line other files are interactive i e the program will prompt users for each parameter one line at a time during execution The Static means it does not require a run time FORTRAN library the default requires 11bF77 s0 x libraries The new or any versions after that 1997 or later will not require files and data being in the current directory to run Other versions may also be compiled at a special request to mzhang cshl org Current Protocols in Bioinformatics BASIC PROTOCOL 2 Finding Genes 4 2 3 Using MZEF to Find Internal Coding Exons 4 2 4 A FASTA file APPENDIX 1B with no more than 80 characters per line that contains the DNA sequence in which one wishes to identify the exons MZEF can only take the standard DNA RNA character symbols either in capital or lower case letters ambiguous IUPAC symbols APPENDIX 1A will be converted to the standard symbols by a random draw e g N will be converted into A C G T with equal probability The example used in the following is a 19 kb human genomic DNA sequence containing the serum albumin ALB gene File name m12523 fasta GenBank accession number M12523 gi 178343 Minghetti et al 1986 The sequence may also be found on the Current Protocols in Bioinformatics Web site at http www3 interscience wiley com c_p cpbi_sampledatafiles htm This gene has an alternative last exon the CDS an
125. and in other model organism genome sequences as well requires the devel opment and application of robust computational methods some of which are listed in Table 4 1 1 These methods provide the first best guess not only of the number and position of all genes but of the structure of each individual gene as well These predictions brought under the banner of sequence based annotation help to increase the intrinsic value of genome sequence data found within the public databases REMEMBERING BIOLOGY IN DEDUCING GENE STRUCTURE In considering the problem of gene identification it is important to briefly review the basic biology underlying what will become in essence a mathematical problem Fig 4 1 1 At the DNA level upstream of a given eukaryotic gene there are promoters and other regulatory elements that control the transcription of that gene The gene itself is discontinuous being comprised of both introns and exons Once this stretch of DNA is transcribed into an RNA molecule both ends of the RNA are modified with the 5 end being capped and a poly A signal being placed at the 3 end The RNA molecule reaches maturity when the introns are spliced out based on short consensus sequences found both at the intron exon boundaries and within the introns themselves Once splicing has occurred and the start and stop codons have been established the mature mRNA is transported through a nuclear pore into the cytoplasm at which point tr
126. and its accuracy in predicting first exons and promoters using a test set of experimentally known first exons is described The performance of FirstEF over a large genomic regions human chromosomes 21 and 22 is also discussed Davuluri R V Grosse I and Zhang M Q Sub mitted FirstEF has been used to perform an initial compu tational annotation of the promoters and the first exons for all 24 human chromosomes Visit http genemap med ohio state edu for accessing the annotations Internet Resources http rulai cshl org tools FirstEF The FirstEF Web based version http genemap med ohio state edu The Bioinformatics Unit of the Human Cancer Ge netics Program at The Ohio State University The First Exon genome browser is available from this site http www ncbi nih gov BLAST The BLAST and MEGABLAST homepage at the NCBI See UNITS 3 3 amp 3 4 for more information http pbil univ lyon1 fr sim4 htm The SIM4 Web site See Florea et al 1998 for more information http genes mit edu GENSCAN html The GENSCAN server at MIT http www cshl edu mzhanglab or http rulai cshl edu The Zhang Laboratory Computational Biology and Bioinformatics website A link to MZEF is available through this site http www wormbase org db seq frend The Sequence Feature Renderer home page Figure 4 7 3 was created using this tool Contributed by Ramana V Davuluri Ohio State University Columbus Ohio Current
127. ans sequences dna WormBaseFTP site http www repeatmasker org RepeatModeler html RECON site the newest version of RECON is available from the RepeatMasker http www bioperl org BioPerl Web site Current Protocols in Bioinformatics
128. anslation can take place While the process of moving from DNA to protein is obviously more complex in eukaryotes than in prokaryotes the mere fact that it can be described in its entirety in eukaryotes would lead one to believe that predictions can confidently be made as to the exact positions of introns and exons Unfortunately the signals that control the process of moving from the DNA level to the protein level are not very well defined precluding their use as foolproof indicators of gene structure For example upwards of 70 of promoter regions contain a TATA box but because the remainder do not the presence or absence of a TATA box in and of itself cannot be used to assess whether a region is a Contributed by Andreas D Baxevanis Current Protocols in Bioinformatics 2004 4 1 1 4 1 9 Copyright 2004 by John Wiley amp Sons Inc UNIT 4 1 Finding Genes 4 1 1 Supplement 6 Table 4 1 1 Web Sites for Common Gene Finding Programs Web site URL Banbury Cross http igs server cnrs mrs fr igs banbury The Encyclopedia of DNA elements ENCODE http www genome gov encode FGENES http genomic sanger ac uk gf ef shtml FirstEF http rulai cshl org tools FirstEF geneid UNIT 4 3 http www1 imim es geneid html GeneMachine http research nhgri nih gov genemachine GeneMark UNITS 4 5 amp 4 6 GeneParser http opal biology gatech edu GeneMark http beagle colorado edu eesnyder GeneParser himl
129. as PATSCAN ihttp www unix mcs anl gov compbiolPatScan HTML can be passed into this version of geneid Current Protocols in Bioinformatics which then predicts genes with in frame TGA codons only when an appropriate SECIS element has been predicted at the appropriate location A prototype of this tool has been used to scan for potential selenoproteins in Drosophila melanogaster and Takifugu rubripes See Castellano et al 2001 2004 for further details Literature Cited Abril J F and Guig6 R 2000 gff2ps Visualizing genomic annotations Bioinformatics 16 743 744 Altschul S F Gish W Miller W Myers E W and Lipman D J 1990 Basic local alignment search tool J Mol Biol 215 403 410 Aury J M Jaillon O Duret L Noel B Jubin C Porcel B M Segurens B Daubin V Anthouard V Aiach N Arnaiz O Billaut A Beisson J Blanc I Bouhouche K Camara F Duharcourt S Guig R Gogendeau D Katinka M Keller A M Kissmehl R Klotz C Koll F Le Mouel A Lepere G Malinsky S Nowacki M Nowak J K Plattner H Poulain J Ruiz F Ser rano V Zagulski M Dessen P Betermier M Weissenbach J Scarpelli C Schachter V Sperling L Meyer E Cohen J and Wincker P 2006 Global trends of whole genome duplications revealed by the ciliate Paramecium tetraurelia Nature 444 171 178 Birney E and Durbin R 2000 Using GeneWise in the
130. aster and other species in the directory param By default geneid produces results in plain text which are sent to the standard output Unix terminal These can then be redirected to a file or another program In particular they can serve as input to programs producing graphical visualization of genomic annotations e g gff2ps or apollo UNIT 9 5 or genome browsers that display such information on the Web e g UCSC genome browser UNIT 1 4 ENSEMBL UNIT 1 15 etc The interaction between geneid and these systems is shown in Basic Protocol 2 2 Examine the results returned by geneid By default geneid output consists of a series of genes predicted along the input sequence geneid uses its own default output format Other more standard formats can be specified via command line options see steps 5 6 and 7 Predicted genes are described as lists of potential coding exons For sequence examplel geneid predicts an eight exon gene see Fig 4 3 1 for plain text output and Figs 4 3 5 4 3 6 and 4 3 7 for graphical representations Each exon is defined by a start signal start codon or acceptor site an end signal donor site or stop codon the strand and the frame Each exon as well as each signal is assigned a score The score depends on the scores of the defining sites and on the nucleotide composition of the exon sequence measuring the likelihood of the exon see Background Information The score of a gene is the sum of the
131. atMasker locally one must obtain RepeatMasker cross_match and correct repetitive libraries from Repbase Update as detailed below It is also possible to run RepeatMasker with WU BLAST see Alternate Protocol for faster processing Current Protocols in Bioinformatics NOTE Investigators unfamiliar with the Unix environment should read APPENDIX ic and APPENDIX 1D Necessary Resources Hardware Any Unix or Linux workstation Software RepeatMasker The software is now licensed under the Open Source License v 2 1 and can be downloaded from hitp www repeatmasker org RMDownload html cross_match This software is part of the Phred Phrap Consed hitp www phrap org consed consed html howToGet also see UNIT 11 2 package It is also free for academic use Write to Phil Green phg u washington edu and include the following information in the message a name b an acknowledgement of agreement to observe the licensing conditions described on the above Web site state that cross_match is desired c institution department d e mail address for all future correspondence ideally e mail should be received through a Unix computer running a generic mail program since several of the programs are sent as unencoded files which may be corrupted by some mail programs Note that it takes up to 2 weeks for a license application to be processed Repbase Update This database http www girinst org Jurka 2001 manages a Fil
132. ay miss more real exons than other gene finders This is particularly true for short exons Compared to other programs the problem is more relevant when analyzing single gene sequences The coding fraction of initial exons is often very short and geneid may not resolve it well missing it completely or extending it into a longer internal exon When analyzing sequences coding for only one gene the authors recommend that a gene model see below be used which forces the prediction of a single gene in the query genomic sequence This single gene can also be forced to be complete thus necessarily starting with a first exon and ending with a terminal exon see geneid manual Exhaustive analysis data not shown indicates that when using this option the accuracy of geneid predictions in single gene sequences compares favorably to that of other gene finders In general for large genomic sequences encoding multiple genes the overall accuracy of default geneid is comparable if not superior to that of the most accurate existing tools offering a better balance between specificity and sensitivity see the geneid Web page for a discussion of accuracy Gene and Exon Scores Gene and exon scores have a probabilistic interpretation within geneid see Background Information Thus although the authors have not studied exhaustively the false positive rate of exon predictions as a function of the score as a rule of thumb the higher the score of an exon the
133. bacterial genomes This heuristic approach can be used for sequences as small as 10 kbp making it especially useful in the analysis of small genomes such as viruses and phages for which there are not enough genes to build accurate models of higher order After creating a model from the sequence data GeneMark hmm and or GeneMark will use the model just created to analyze the sequence Necessary Resources Hardware A personal computer or workstation with Web access Software A Web browser Files A single sequence in FASTA format APPENDIX 1B The sample sequence example fna which contains region 1 to 50 000 from Escherichia coli K12 used to illustrate this protocol can be downloaded from the Current Protocols Web site http www3 interscience wiley com c_p cpbi_sample datafiles htm 1 Via a Web browser connect to http opal biology gatech edu GeneMark sheuristic_ hmm2 cgi In the Input Sequence section paste an input sequence into the Sequence box area or alternatively click on Browse next to the Sequence File Upload box to upload the input sequence file from a local drive The Sequence File Upload option is more powerful since the copy and paste method imposes a limit on the length of the sequence If the sequence has a FASTA APPENDIX 1B title line e g gt Sequence name this name will be assigned to the sequence in the output unless the user gives a name in the Sequence Title text area For the purpose of a
134. be down loaded by anonymous FTP from tp genome imim es pub software gff_tools gff2ps How to obtain apollo The apollo Web page contains the information required to download and install the program http www fruitfly org annot apollo The default apollo configuration deals with geneid GFF files successfully However to reproduce the graphical representation in Figure 4 3 6 one will need to add some lines to the file ensj tiers This file contains the display specifications for different gene features Therefore open the file with a Unix editor such as pico joe or emacs type the following lines at the end of the file save the resulting file and close the editor Tier tiername geneidv1 2 Type label geneid_v1 2 tiername geneidv1 2 datatype geneid_v1 2 glyph DrawableGeneFeatureSet color 255 204 0 column GENOMIC_LENGTH column GENOMIC_RANGE column SCORE In addition the style attributes that must be set in the preferences file are FeatureBackgroundColor black EdgematchColor green EdgematchWidth DN CoordBackgroundColor black CoordForegroundColor yellow Draw3D true GUIDELINES FOR UNDERSTANDING RESULTS Despite significant advances in the field of computational gene prediction current gene finding methods are far from being able to accurately predict the exonic structure of the genes encoded in large genomic sequences for a recent exhaustive evaluation of gene
135. be superior Of course there are also other nonparametric methods that are beyond the scope of this unit Discriminant analysis can be done equally well by neural networks or machine learning approaches where the decision boundary or the distribution parameters are estimated by iteration algorithms Bishop 1996 here the multivariate statistical approach for its analytical clarity is the focus Feature Variables Used in MZEF If fa is some frequency found in class A the author defines a preference for A versus B say exons versus pseudoexons to be the ratio pap fa fa fg It is clear that if fa lt lt fp the preference for A would be close to zero if fa fg the preference for A would be 4 no preference There are nine feature variables used in MZEF and they are computed for high or low 0 48 being the cutoff G C query sequences separately Suppose f_exon and f_intron are frequencies for 6mers or 3mers in the exon and intron regions pre computed from the training data then these 9 feature variables computed on the flight are 1 Exon_length score x log bp 2 Intron exon_transition score x average intron_preference to the left exon_preference to the right sum Of Pintronexon OVEF all overlapping 6mers in the 54 bp window to the left of 3 ss sum Of Pgxon intron OVEF all overlapping 6mers in the 54 bp window to the right of 3 ss 49 3 Branch site score x maximum log likelihood branch si
136. bols are ignored and all ambiguous letters other than the symbols of the four nucleotides assuming that they occur rarely are replaced with C This minimizes the chance of the possible creation of a false start or stop codon 2 Scroll down the page and select the name of the species of interest from the Species pull down menu which will result in the selection of the corresponding statistical model By means of the two check boxes below the Species pull down menu the user may choose either the Typical model the Atypical model or both Prokaryotic Gene Prediction Using sa GeneMark and Choosing the correct species name is essential to obtaining meaningful results since wrong GeneMark hmm statistical models may totally corrupt the results of gene prediction 4 5 8 Supplement 1 Current Protocols in Bioinformatics amm 40 GeneMark hmm predicti chorichia col K12 Region 1 50000 Order 2 Window 96 Step 12 c25 Order 2 Window 96 10 osphh s t 9 p v oe oA Ne 1 _ _ h ae Oe eT g 5 s 7 PE EEN CSEE WORN SF CEE PO eg DEN PETE Y E i T R E a 10 8800 9200 9600 10000 10400 OSfer 1s 1 teens eee eh y eee o 8 et M foii Steg gy Direct Sequence OSF m s ari p eat imm _i t mm t Pe EREN TEET EEN TEE PP N PORNS OR EEA S F BOR OO DORN TAA NS g 10 8800 9200 9600 10000 10400 5 5 aon wee tran T pet Ag mR Tarar ado i DAPA D 5 se ee ee BEN TE eee I 10 8800 9200 9600 10000 10400 os tome el
137. bove the central scale bar showing the nucleotide coordinates and exons predicted in the reverse strand are displayed below Exons belonging to the same gene are joined together by a line Zoom in using the x10 and x2 buttons and zoom out using the x l and x 5 buttons Use the scroll bar to move along the sequence The detail panel underneath the main panel shows information about any feature or set of features selected The left hand panel shows the type and color of the feature the name the range and the score The right hand panel shows more information about each individual exon genomic range genomic length and score Selecting an individual exon in the main window causes the exon to be selected in the right hand panel for an easier identification and vice versa Visualization using the UCSC genome browser Ic 2c 3c Ac In order to import the geneid GFF output in the UCSC genome browser it is necessary to first locate the correct chromosomic location in the human genome for the sequence examplel fa Load the UCSC genome browser Web page and click over the Genome Browser link Select the human genome Then introduce the coordinates of this region in the Position box chr21 13 903 812 13 935 812 and press the Submit button The main UCSC genome window will appear The basic unit of information in the UCSC genome browser is the track Each track is a graphical line of annotations in the image Graphical features are associa
138. by linguistic methods Genomics 23 540 551 Fickett J W 1982 Recognition of protein coding regions in DNA sequences Nucl Acids Res 10 5303 5318 Fickett J W and Tung C S 1992 Assessment of protein coding measures Nucl Acids Res 20 6441 6450 Gelfand M S 1990 Computer prediction of the exon intron structure of mammalian pre mRNAs Nucl Acids Res 18 5865 5869 Guigo R Knudsen S Drake N and Smith T 1992 Prediction of gene structure J Mol Biol 226 141 157 Henikoff S and Henikoff J 1991 Automated as sembly of protein blocks for database searching Nucl Acids Res 19 6565 6572 Hutchinson G B and Hayden M R 1992 The pre diction of exons through an analysis of spliceable open reading frames Nucl Acids Res 20 3453 3462 Hyatt D and Uberbacher E C 2002 Computa tional DNA sequence analysis and annotation In Genomic Technologies Present and Future D J Galas and S J McCormack eds pp 345 374 Caister Academic Press Norfolk U K Mani G S 1992 Long range correlations in DNA and the coding regions J Theor Biol 158 447 464 Mural R J Einstein J R Guan X Mann R C and Uberbacher E C 1992 An artificial intelli gence approach to DNA sequence feature recog nition Trends Biotech 10 67 69 Snyder E E and Stormo G D 1993 Identification of coding regions in genomic DNA sequences An application of dynamic programming and neural networks Nu
139. can Genas GralEXP CpG Isiands Rapaathtasker Repeats IRNASCAN SE RNA Ganas Gral Dacpairs E PCR STSS Gene Varant Strand Exons Begin End StartCodon _ StopCodon EvidenceBegin EvidenceEnd 1 1040 36422 75055 30147 36040 36422 2 3836 35979 4031 35664 3896 7 f i 3998 35979 4031 35664 3096 f 1836 35979 4031 35004 3096 1 Figure 4 9 5 Java Pipeline Viewer This black and white facsimile of the figure is intended only as a placeholder for full color version of figure go to http www interscience wiley com c_p colorfigures htm GUIDELINES FOR UNDERSTANDING RESULTS Major Gene Modeling Issues in GrailEXP Gawain Alternative splicing recognition Gawain currently looks for a very specific case of alternative splicing It looks for fully determined EST evidence that indicates the insertion or deletion of an exon In other words if one gene model contains exons A B and C and another contains only exons A and C then the program identifies this case as an alternative splice Except in the case of repeating gene regions which can be incorrectly reported as regions with a lot of alternative splices Gawain identifies inserted omitted exons with absolute certainty However it cannot currently identify other types of less obvious alternative splicing It is not a trivial matter to distinguish between an unspliced product that has slipped into the EST database an error in the genomic alignment and a genuine cas
140. ce site at its ending position all in the same reading frame The exon candidates are catego rized into clusters with the highest scoring exon in each cluster clearly indicated These best exons in each cluster are traditionally referred to as GRAIL exons rather than can didates Perceval identifies and scores all potential splice sites within the sequence using neural networks trained for recognizing start codons stop codons AAG and YAG acceptors and GT donors respectively Low scoring splice sites are discarded All possible exon candidates that can be constructed with the remaining splice sites are then scored for their coding potential Here too the low scoring candidates are dis carded The remaining exon candidates are evaluated by another neural network which is fed the splice site and coding potential scores as well as the GC content Once again only the high scoring exon candidates are retained as the final set This set of exon candidates is then organized into clusters of overlapping exons Each cluster is filtered for repetitive elements using NCBI s BLAST program UNITS 3 3 amp 3 4 Non exonic regions of the sequence are substituted with N s and a BLAST search of this sequence is run against a database of repetitive elements If a repetitive element is determined to have a sig nificant overlap 10 of the exon if overlap ping an edge 50 if embedded inside the exon with an exon candidate th
141. cedure is beyond the scope of this unit For necessary resources see Alternate Protocol 1 To run the Perl script change to the directory N SCAN bin Next run the N SCAN pipeline by entering the following command on a single line Nscan_driver pl masked target sequence configuration file where masked target sequence is the sequence to be annotated see Alternate Protocol 1 for instructions on file format and configuration file contains the full paths to all programs An example configuration file can be created by running Nscan_driver pl config gt configfile The resulting config file must be edited to reflect the local configuration The output files are created in the current directory by default To specify an alternate directory include d output directory The Nscan_driver p1 script will generate files whose names consist of the target sequence file name with the following extensions masked align and gtf Ifany of these files already exists the script will print a warning and stop running Delete or move the existing files before rerunning The Nscan_driver pl program will mask the target sequence create a align format file using Blastz and the file format conversion scripts and run N SCAN If the sequence is already masked the command can be run as follows Nscan_driver pl nomask target sequence configuration file Ifa align file already exists for the target sequence run Nscan_driver pl bla
142. ch prediction these models will use varying numbers of bases for each prediction In some contexts they will use 5 bases while in others they might use 26 bases and yet in other cases they may use lt 4 bases This allows IMMs to be sensitive to how com mon a particular oligomer is in a given genome In a given genome many 5 mers might occur rarely and should not be used for prediction here the IMM will fall back ona shorter Markov chain On the other hand certain 8 mers may occur very frequently and for those the IMM can use this longer context and make a better prediction In addition the IMM can combine the evidence from the 8 order Markov chain and the 5 order chain in such cases Thus it has all the information available to a 5 order chain plus additional information It is also worth noting that both IMMs and 5 order Markov chains should outperform methods based on codon usage statistics GlimmerM uses the same IMM algorithm as the one described by Salzberg et al 1998a in the original Glimmer publication IMMs form the basis of the Glimmer system which finds genes in prokaryotes bacteria archaea viruses and in a few very simple eukaryotes T brucei Glimmer correctly identifies 99 of the genes in bacteria without any human intervention and with a very limited number of false positives Since its introduction it has been used as the gene finder for B burgdorferi Fraser et al 1997 T pallidum Fraser
143. chosen for the start s can be ulted in false ata predator npertea Halaria train trainli_o1 Figure 4 4 3 An example log file generated by trainGlimmerM 4 4 8 Current Protocols in Bioinformatics Table 4 4 2 Parameters for the Configuration File Called config file see Fig 4 4 2 Line no in the config file Parameter 1 Minimum intron length 2 Maximum intron length 3 Minimum exon length 4 Maximum exon length 5 Maximum gene length 6 A flag indicating if decision trees were used for computing the acceptor site scores 7 A flag indicating if decision trees were used for computing the donor site scores 8 Acceptor site threshold 9 Donor site threshold 10 Length of the filter window for the acceptor sites 11 Length of the filter window for the donor sites 12 Acceptor site threshold when filtering is used 13 Donor site threshold when filtering is used 14 A flag indicating if start codon modeling was used 15 Start codon threshold attributes Each edge extending from an internal node of the tree represents one of the possible alternatives of courses of action available at that point So depending on the outcome of the test different paths in the tree are followed down to the tree leaves that carry the class names into which the objects are classified 6 Change all of the other parameters with the exception of parameters 6 7 and 14 shown in Table 4 4 2 by opening the config file witha
144. ckling it it is important for investigators to appreciate when and how each particular method should be applied A recurring theme in this chapter will be the fact that each method will perform differently depending on the nature of the data Put another way while one method may be best for human finished sequence another may be better for sequences whether they be finished or unfinished from another organism The reader will also notice that the various methods in this chapter produce different types of results in some cases lists of putative exons are returned but these exons are not in a genomic context in other cases complete gene structures are predicted but possibly at a cost of less reliable individual exon predictions Returning to the cautionary note that different methods will perform better or worse depending on the system being examined it becomes important to be able to quantify the performance of each of these algorithms Several studies have systematically examined the rigor of these methods using a variety of test data sets Burset and Guig6 1996 Claverie 1997a Snyder and Stormo 1997 Stormo 2000 Rogic et al 2001 Before discussing the results of these studies some definition of terms is in order Current Protocols in Bioinformatics Finding Genes 4 1 3 Supplement 6 An Overview of Gene Identification 4 1 4 Supplement 6 TP FP TN FN TP FN JTN
145. cl Acids Res 21 607 613 Solovyev V V Salamov A A and Lawrence C B 1994 Predicting internal exons by oligonu cleotide composition and discriminant analysis of spliceable open reading frames Nucl Acids Res 22 5156 5163 Staden R 1984 Measurements of the effects that coding for a protein has on a DNA sequence and their use for finding genes Nucl Acids Res 12 505 519 Uberbacher E C and Mural R J 1991 Locating protein coding regions in human DNA se quences by a multiple sensor neural network approach Proc Natl Acad Sci U S A 88 11261 11265 Xu Y Mural R Shah M and Uberbacher E 1994a Recognizing exons in genomic sequence using GRAIL Il Jn Genetic Engineering Princi ples and Methods J K Setlow ed vol 15 pp 241 253 Plenum New York Xu Y Mural R J and Uberbacher E C 1994b Constructing gene models from accurately pre dicted exons An application of dynamic pro gramming CABIJOS 10 613 623 Key References Altschul S F Gish W Miller W Myers E W and Lipman D J 1990 Basic local alignment search tool J Mol Biol 215 403 410 Antequera F and Bird A 1993 Number of CpG islands and genes in human and mouse Proc Natl Acad Sci U S A 90 11995 11999 Bairoch A 1993 The PROSITE dictionary of sites and patterns in proteins its current status Nucl Acids Res 21 3097 3 103 Bairoch A and Boeckman B 1993 The SWISS PROT protein sequence data
146. col Need for organism specific training GlimmerM trained for rice produces clearly superior gene models for rice sequences than GlimmerM trained for Arabidopsis Pertea and Salzberg 2002 A more thorough evaluation was done by running GlimmerM on 42 genes with EST or protein sequence homology that were extracted from rice BACs in the TIGR databases www tigrorg tdb rice The Arabi dopsis version of GlimmerM detected only 65 of the coding sequences of the genes with a specificity of 86 while the rice version detected 93 of the coding nucleotides with 90 specificity More recent experience with T parva further supports this observation Us ing the P falciparum version of the system produces vastly inferior predictions to a version trained on T parva genes when analyzing T parva data Current Protocols in Bioinformatics Although the currently trained versions of GlimmerM should work well on closely related organisms the user should use the suggested training procedure to re train the system for other eukaryotic organisms in order to improve the accuracy of the gene prediction If very few genes are available however then the best course is to use a version of GlimmerM trained on the most closely related organism For spe cies on which GlimmerM is already trained further improved performance by re training the system as the amount of DNA sequences and continued growth of the number of vali dated genes from these spec
147. d training procedure and can be used with no prior knowledge of any protein However the current version of the program does not find introns which may occur though rarely in viral genes USING WEB INTERFACE GeneMark hmm FOR EUKARYOTIC GENE PREDICTION The Web site shown in Figure 4 6 1 is an interface for the Eukaryotic GeneMark hmm program which is run on an IBM RS 6000 server at the School of Biology of the Georgia Institute of Technology The gene prediction results are reported as a list of exon coordinates Fig 4 6 2 Optionally graphical output and a list of predicted protein sequences can be produced This protocol describes GeneMark hmm version 2 2 and GeneMark version 2 4 Necessary Resources Hardware A personal computer or workstation with Web access Software A Web browser e g Netscape Communicator or Microsoft Internet Explorer Files A single sequence in FASTA format see APPENDIX 1B and Pearson 1990 The sample sequence DNA sequence of complete human serum albumin ALB gene GenBank accession no M12523 used to illustrate this protocol can be Contributed by Mark Borodovsky Alex Lomsadze Nikolai Ivanov and Ryan Mills Current Protocols in Bioinformatics 2003 4 6 1 4 6 12 Copyright 2003 by John Wiley amp Sons Inc UNIT 4 6 BASIC PROTOCOL Finding Genes 4 6 1 Supplement 1 Eukaryotic GeneMark hawn Microsoft Internet Ceplorer fe foe Yew Apote joos iko gt
148. d on the browser or will be sent to the E mail address provided 5 Interpret the text output The GeneMark text output Fig 4 5 2 contains the following sections a Report header Each report generated by GeneMark has a header confirming the parameters selected by the user in step 2 and indicating the name and order of the statistical model matrix used in the analysis b List of Open Reading Frames The sequence positions designatedas Left endandRight end define the boundaries of a predicted open reading frame relative to the sequence start 5 end of the direct strand DNA Strand indicates in which strand the coding region is located direct or comple ment and Coding Frame indicates the absolute reading frame The Avg Prob column denotes the average coding potential over the indicated sequence range GeneMark does not indicate if an ORF extends beyond the limits of the sequence provided so ORF end positions at 1 2 3 and at L 2 L 1 L where L is the sequence length may indicate that the ORF observed is just part of an ORF The value shown in Start Prob column is the likelihood that the start of the open reading frame is the actual start For possible gene starts located closer than the window length to the sequence ends this value is not calculated If an RBS model is specified the program will predict putative RBS sites The RBS Prob value is a score indicating the likelihood of a particular oligomer usually hexamer
149. d position and length of each predicted exon Noncoding exons are called UTR for untranslated region protein coding exons are CDS The first coding exon usually contains an untranslated region so this exon has both types it is listed as UTR CDS In a later version 3 UTRs will be included in the output Below the exon table you will find the predicted protein Click on the link to the NCBI BLAST page to submit this protein or the corresponding transcript to NCBI s BLAST server UNITS 3 3 amp 3 4 This is useful for checking whether the predicted gene is a known gene or is homologous to known genes If the UCSC genome browser UNIT 14 lists the target species of this submission a second link will be shown Click on this link to align your sequence to its genome in the genome browser This allows you to see neighboring genes as well as many other gene predictions and annotation tracks in the region To see the predicted transcript click on the Transcript button below the protein sequence Current Protocols in Bioinformatics 10 Download output Gene predictions can be downloaded as transcript sequences protein sequences or GTF files GTF stands for gene transfer format one of the standard formats for genome annotation GTF is derived from the GFF format developed at the Sanger Center This file can be examined manually to obtain exact coordinates for one or two genes of particular interest but it is intended primarily for automate
150. d processing Each line of the GTF has the following fields separated by tabs seqname source feature start end score strand frame attributes The feature field consists of one the words start_codon stop_codon 5UTR or CDS The attributes field contains the word gene_id followed by an automatically gener ated id symbol then a semicolon the word transcript _id and another automatically generated symbol Coding regions and other features with the same transcript id belong to the same transcript The inclusion of both gene_id and transcript _id allows for alternative splices of a single gene but that feature is not currently used by N SCAN More information on the GTF format can be found at http mblab wustl edu GTF22 html To download all predictions click one of the links at the top of the page next to the gene overview table To download the sequence or coordinates for any one gene click on the appropriate link next to its exon table 11 Manage your submissions Click on My Submissions at the top right corner of the Submission or N SCAN input page This will show an overview of all your submissions Fig 4 8 5 From here you can access the Submission page of each of your jobs by clicking on its Submission link The overview also lists the status of your running jobs and when they are complete a link to the GTF Press the Delete button if you no longer need the results N SCAN Twinscan Gene Predictor
151. d sites exons This measure characterizes how many sites are overpredicted e g the higher the specificity the lower the overprediction rate and the more reliable the predictions On the other hand sensitivity is defined as the number of true predicted sites exons over the whole number of annotated sites exons The higher the sensitivity the lower the chance that a true site exon has not been predicted This measure characterizes how well the program recognizes the actual sites or exons The two measures of accuracy specificity and sensitivity are inversely related with one in creasing when the other decreases The accuracy of two programs may be considered similar if the increase decrease of specificity is compensated by the decrease in crease in the sensitivity Therefore when comparing the performance of different gene prediction programs it is important to take into account both measures of accuracy GeneMark hmm produces sufficiently accurate prediction results with high sensitivity and specificity for a wide range of eukaryotic organisms These results are achieved by tuning up the organism specific models see Background Information discussion of Model Construction COMMENTARY Background Information Hidden Markov model framework The GeneMark hmm algorithm Lukashin and Borodovsky 1998 M Borodovsky and A V Lukashin unpub observ was designed to improve gene prediction quality in terms of finding exact
152. d the informant sequence s in FASTA format APPENDIX 1B The informant sequences can be shotgun reads or assemblies A parameter file For organisms available on the website parameter files are included in the parameters directory of the N SCAN distribution See Commentary for a discussion of parameter estimation and substitution 1 Mask repeats in the target sequence by replacing them with N s N SCAN s accuracy is somewhat improved by replacing sequences that are unlikely to contain genes of interest with N s Such sequences include mobile repetitive elements and functional RNA genes such as tRNAs The effect of repeat masking is greater the more repetitive the genome but even in relatively nonrepetitive genomes like that of C elegans masking repeats yields small improvements If RepeatMasker is installed repeat masking is achieved by the command RepeatMasker options target sequence file This command creates several files including a masked sequence This sequence has the same name as the input sequence followed by masked Use this file in all the commands described below To view the current masking options type RepeatMasker with no arguments If the species or clade of interest is not listed it is possible to specify the name of a repeat library file directly For more information on how to run RepeatMasker see UNIT 4 10 or run RepeatMasker h There is also a RepeatMasker Web server at http www repeatmasker o
153. d to the sequence in the output unless the user gave a name in the Sequence Title text area For the purpose of analysis all nonalphabet symbols are ignored and all ambiguous letters other than the symbols of the four nucleotides assuming that they occur rarely are replaced with C 2 Scroll down the page to the Running Options and select the option Use Eukaryotic Virus Version 3 Scroll further down the page and set the Output Options The user may request the graphical output An E mail address is required for sending text output for sequences longer than 1 Mbp or if graphical output is requested Current Protocols in Bioinformatics 4 After completing the above entries click the Start GeneMarkS button The results will be depicted on the browser or will be sent to the E mail address provided 5 Interpret the text output The text output from GeneMarkS is identical to that of GeneMark hmm see UNIT 4 5 Alternate Protocol 1 6 Interpret the graphical output The graphical output from GeneMarksS is identical to that of GeneMark hmm see UNIT 4 5 Alternate Protocol 1 when using just one model GUIDELINES FOR UNDERSTANDING RESULTS The accuracy of prediction is described in terms of sensitivity and specificity both defined for each signal site translational start and stop donor and acceptor and for each exon as a whole Specificity is defined as the number of true predicted sites exons over the number of all predicte
154. d uses a probability model to combine this information with information from patterns in the target DNA sequence The model exploits the fact that features such as introns UTRs coding sequence and splice sites all exhibit characteristic patterns in the target sequence and all evolve under distinct selective pressures leaving an imprint on local patterns of conservation Since N SCAN uses a combined model it can predict genes that are strongly indicated by the target DNA sequence even if they are not well conserved in the informant genome s N SCAN can be run either through a Web browser pointed at the N SCAN server http mblab wustl edu nscan submit or through the command line on a local computer The Web server is recommended for all users except those with substantial bioinformatics experience and the need to 1 process gt 1 Mb per day on a sustained basis 2 use N SCAN outside the supported clades or 3 use N SCAN on proprietary sequences that cannot be stored on the server To use the Web server the user is required to register A personal account is created so that the user can track the progress and results of all gene predicion submissions The Basic Protocol describes using the Web interface of N SCAN to find gene structures exons coding sequences and protein sequences The Support Protocol describes how to obtain and install N SCAN software on a local computer running Linux Alternate Protocol 1 describes how to run N SCAN o
155. de Position Figure 4 5 3 The graphical output from the GeneMark program for a region of the example sequence example fna The six different panels represent the six possible reading frames three each on the direct and reverse strands Finding Genes 4 5 5 Current Protocols in Bioinformatics Supplement 1 ALTERNATE PROTOCOL 1 Prokaryotic Gene Prediction Using GeneMark and GeneMark hmm 4 5 6 Supplement 1 d Detection of Possible Frameshifts GeneMark indicates possible frameshifts in protein coding regions A frameshift in the graphical output produces a switch of the coding potential graph from one panel to another panel related to the same DNA strand This situation occurs when there is an insertion or deletion of one or several nucleotides not a multiple of three in a coding region The table indicates the frame in which the codon region started the frame in which the coding region continues and the approximate location of the frameshift the precision of which is determined by the Step Size parameter used sample not shown 6 Interpret the graphical output The GeneMark graphical output Fig 4 5 3 depicts the coding potential in the six possible reading frames three each on the direct and reverse strands An unbroken horizontal line at the 0 5 level indicates an open reading frame ORF Large vertical ticks above the 0 5 level indicate ATG codons while small vertical ticks represent G
156. dict full exonic structures of vertebrate genes in anonymous DNA sequences geneid was designed follow ing a simple hierarchical structure first gene defining signals were predicted and scored using weight matrices Next potential exons were constructed from these sites and their coding potential was scored as a function of several coding statistics such as hexamer com position whose coefficients were estimated by a neural network Finally the optimal scoring gene prediction was assembled from the best exons by performing an exhaustive search of the space of possible gene assemblies ranked according to a score obtained through a com plex function of the score of the assembled exons Roderic Guig6 Steen Knudsen and Neil Drake in the Temple F Smith group con Current Protocols in Bioinformatics Finding Genes 4 3 23 Supplement 18 Using geneid to Identify Genes 4 3 24 Supplement 18 memory efficient It was more accurate than version v1 0 parameter files were developed for a larger number of species and more exten sive documentation was supplied The latest version of geneid v1 2 has been released in 2004 Additional parameter files and other new features were implemented over the platform of the previous version to support compara tive gene prediction The code in the geneid v1 x versions is mostly by Enrique Blanco and Roderic Guig6 with contributions from Moi ses Burset and Xavier Messeguer
157. ditionally it is not mandatory to provide an exon candidate input file to Galahad In absence of exon input Galahad performs alignment of the entire se quence with the search database In such cases it is recommended that a repeat masked se quence be provided as input otherwise the processing time would be significant due to the large number of alignments that would be iden tified Overall Galahad provides the advantage of speed and flexibility compared to other alignment systems Gawain Gawain Gene Assemblies with Alignment Information the GrailEXP gene assembly program builds gene models based on GRAIL Exon Candidates and or GrailEXP genomic alignments A GrailEXP gene model incorporates the following components a 5 untranslated region a coding region a 3 un translated region exons introns a polyadeny lation poly A site a promoter and all refer ence sequence evidence consistent with the gene model The program also generates the mRNA and protein translation for each gene model New capabilities in this version of GrailEXP include recognition of alternative splicing and clustering of all ESTs cDNAs that support a particular gene model Gawain can construct gene models from either exon candi dates or alignments or both Gawain first clusters the genomic align ments and each alignment is assigned a frame using a recursive frame scoring function that considers the GRAIL coding score for each potential fr
158. dow column Note that all promoters listed in this example have a length of 570 nt because the promoter QDF was trained on a promoter of length 570 nt Examine the a posteriori probabilities of promoter exon and donor which are labeled P promoter P exon and P donor respectively The boundaries of Exon are the transcription start sight left boundary and the donor site right boundary The higher the values of the probabilities the higher the chance that the corresponding predictions are real A value of P exon 1 means that the first exon prediction is 100 correct whereas a value of P exon 0 means that the first exon prediction is 100 incorrect P exon values 20 5 are considered significant Determine if the predicted first exon is CpG related If the predicted first exon is CpG related the boundaries of the corresponding CpG Window of length 201 are presented Otherwise the CpG Window entry will read Non CpG related See Background Information for further discussion of CpG win dows Observe the Rank of each first exon prediction within each cluster Combine the resulting predictions with other annotations see Support Protocol Current Protocols in Bioinformatics Finding Genes 4 7 3 Supplement 1 Application of FirstEF to Find Promoters and First Exons in the Human Genome 4 7 4 Supplement 1 D Search f eM M Intermed E agdcee groveded by l Ee g t Yew Ios tep HBk e
159. ducing any prediction The following error message will appear Too many predicted sites Change RSITES parameter or a similar message concerning exon types In order to minimize memory usage geneid makes a guess on the maximum number of sites and exons that will be predicted in a given sequence fragment While for most sequences the guess is correct in some particularly anomalous genomic sequences these numbers are much higher than that guessed The user will need to change the parameters that con trol how these numbers are guessed These parameters are assigned default values in the geneid header file which the user will find at include geneid h within the geneid dis tribution Decrease these values in the header file and recompile geneid see geneid docu mentation for details For instance RSITES is 5 by default so if the message above ap pears change it to 2 then recompile geneid and run it again Users must note that by de creasing these numbers the amount of memory required by geneid may increase substantially geneid produces inconsistent results or crashes after starting or while running In some exceptional cases geneid produces a prediction with inconsistent exon coordi nates or crashes without a warning The au thors believe they have reduced these cases to a minimum which were mostly related to memory management problems If the user en counters these problems please report them to the authors at geneid imim
160. e appears Fig 4 8 1 If your browser allows cookies to be set you will be logged in automatically the next time you go to the N SCAN page 3 Input the target sequence There are two possible ways to submit a sequence to N SCAN a Select Text copy the target sequence and paste it into the input box see Fig 4 8 1 b Load the target sequence from a file by selecting File and clicking the Browse button that appears next A dialog box will pop up Users can select any file on their computer from this dialog box In either case the input must be a nucleotide sequence file consisting of upper or lower case ACGTN If the first line begins with gt it will be treated as a sequence identifier and may contain any characters All other lines may only contain ACGTN in upper lower or mixed case whitespace and numbers This is useful for pasting in files that contain Current Protocols in Bioinformatics Finding Genes 4 8 3 Supplement 20 Using N SCAN or TWINSCAN 4 8 4 Supplement 20 line numbers and other formatting information However if N SCAN detects any other character a warning is issued informing the user that the sequence is not valid The N SCAN Web server only accepts sequences between 500 bp and 2 Mb in length To annotate entire large genomes it is most efficient to download and run N SCAN locally or to contact the authors about collaboration 4 Select the type of sequence masking Before gene
161. e gene prediction in human and mouse Genome Res 13 108 117 Salamov A A and Solovyev V V 2000 Ab initio gene finding in Drosophila genomic DNA Genome Res 10 516 522 Stanke M and Waack S 2003 Gene prediction with a hidden Markov model and a new intron submodel Bioinformatics 19 11215 II225 Stanke M Tzvetkova A and Morgenstern B 2006 AUGUSTUS at EGASP Using EST pro tein and genomic alignments for improved gene prediction in the human genome Genome Biol 7 S11 1 S11 8 van Baren M J and Brent M R 2006 Itera tive gene prediction and pseudogene removal improves genome annotation Genome Res 16 678 685 Wei C and Brent M R 2006 Using ESTs to im prove the accuracy of de novo gene prediction BMC Bioinformatics 1 327 Current Protocols in Bioinformatics GrailEXP and Genome Analysis Pipeline for Genome Annotation Since its inception The Gene Recognition and Analysis Internet Link GRAIL Uber bacher and Mural 1991 Mural et al 1992 has been one of the most widely used systems for locating protein coding genes and several other features of biological interest in DNA sequences and has been used extensively for annotation of human and mouse genomes This unit first describes the use of GrailEXP see Basic Protocol the latest version of this gene finding system from Oak Ridge National Laboratory ORNL where the original GRAIL system was developed GrailEXP provides significant improv
162. e of accuracy GlimmerM predicted the precisely correct structure for 98 out of 113 genes 87 on chromosome 2 of P falciparum and was able to predict partially correct models for 14 others These numbers necessarily leave out any genes on that chromosome for which no independent evidence was available and it is impossible to estimate a false positive rate with the data avail able today The accuracy of GlimmerM has also been evaluated on the model plant Arabidopsis thaliana and it is expected to be similar on Oryza sativa Pertea and Salzberg 2002 Table 4 4 3 presents the authors results on the ARASET data collected by Pavy et al 1999 ARASET contains 74 genomic sequence frag ments with multiple genes in each sequence In total it contains 168 genes and 94 intergenic sequences Note that although the accuracy of the predictions is high at the nucleotide and exon level only 63 out of 168 are predicted perfectly The main reason for the higher accu racy on Plasmodium as compared to Arabidop sis is the difference in gene structure In malaria parasites and many other single celled eu karyotes introns are short and few in number and genes are relatively densely packed along the chromosomes In plants and animals genes have many more introns the introns themselves are much longer and the genes are sparsely distributed along the chromosomes making them much harder to find Critical Parameters and Troubleshooting
163. e of alternative splicing Just because one rogue EST disagrees with the other evidence does not neces sarily mean it is an alternative splice Even if multiple ESTs indicate an alternative splicing it could simply be due to an error in the alignment or in the ESTs In future GrailEXP and versions identification of additional types of alternative splicing will be addressed Genome Analysis Pipeline for Genome Annotation 4 9 8 Supplement 4 Current Protocols in Bioinformatics Identification of 5 and 3 untranslated regions The primary indicator of 5 and 3 untranslated regions is an additional stop codon in the mRNA In such cases the program attempts to find the 5 and 3 untranslated region boundaries The mRNA is analyzed to find the highest scoring run without a stop codon the score being based on the length of the run and the GRAIL coding score It then examines that run to find the highest scoring start codon based on proximity to the beginning of the run which exon it falls in GRAIL start site score and coding noncoding scores around the boundary Gawain correctly identifies the start site in a complete mRNA about 95 of the time Further performance improvement in future versions will be addressed by adding protein similarity search to the system Identification of pseudogenes GrailEXP does not specifically label a pseudogene as such However it does provide some indication of a particular model be
164. e rends to Sane w a Figure 4 7 1 The screen shot of the FirstEF Web page see Internet Resources The README file available with the downloaded version gives a brief description of how to use FirstEF Examples of human genomic sequences analyzed by FirstEF are presented at the Web site Preliminary annotations of the human genome can also be accessed from this Web site 2 Enter desired values for the three parameters listed a The first exon a posteriori probability P exon This value quantifies the probability of finding a true first exon at the predicted location A value of P exon 1 means that the first exon prediction is 100 correct whereas a value of P exon 0 means that the first exon prediction is 100 incorrect b The splice donor a posteriori probability P donor This value quantifies the probability of finding a true splice donor at the predicted location A value of P donor 1 means that the splice donor prediction is 100 correct whereas a value of P donor 0 means that the splice donor prediction is 100 incorrect c The promoter a posteriori probability P promoter This value quantifies the probability of finding a true promoter at the predicted location A value of P promoter 1 means that the promoter prediction is 100 correct whereas a value of P promoter 0 means that the promoter prediction is 100 incorrect These user selected values are the lower boundari
165. e sequence file name or use the Browse button to upload the sequence MZEF can only take the standard DNA RNA character symbols either in capital or lower case letters ambiguous IUPAC symbols APPENDIX 1A will be converted to the standard symbols by a random draw e g N will be converted into A C G T with equal probability For this example cut and paste the contents of the m12523 fasta file into the box Using MZEF to Find Internal Coding Exons 4 2 2 Current Protocols in Bioinformatics Determine which strand should be used Set Strand 1 to analyze the forward Watson strand Set Strand 2 to select the reverse Crick strand For the example shown here select the default value Strand 1 Determine the maximum number of overlapping exons per splice site allowed in the output Enter this integer in the Overlap box See Critical Parameters for further discussion of this parameter For the example shown here select the default value of 0 Determine how likely it is that a randomly picked potential exon AG ORF GT is real Place this value in the Prior box The default value is based on real life training sets and rarely needs to be adjusted See Critical Parameters for further discussion of this parameter For the example shown here select the default value of 0 02 Click the Submit button to have the results displayed on the browser Alternatively have the results sent
166. e site may contain sequencing errors yielding a true site that gets a very low score As previously men tioned Salzberg 1997 maximizing the corre lation coefficient CC is also a poor strategy for choosing the thresholds in part because this statistic gives equal weight to positive and negative examples All the above considerations should be taken into account when setting the threshold In the automated training protocol of GlimmerM see Support Protocol the threshold for calling a sequence a real splice site is chosen by exam ining the trade off between the false positive and negative rates The system creates a sorted list of thresholds adjusting the scoring function so that it will miss 1 2 3 etc true sites For each of the associated false negative values it com putes the false positive rate Typically the false positive rate drops very rapidly at first as the low scoring true sites are excluded from the calculation With each successive removal of a true site the false positive rate falls further but eventually the rate declines more slowly The default threshold is chosen to be the score corresponding to the point at which the false positive rate drops by lt 1 To allow a greater flexibility in setting the signals thresholds the authors training procedure allows the user to consult these false positive and false negative rates and reset the threshold to yield a different tradeoff see Support Proto
167. e the error rate in locating the initialization and termination sites extended margins I to 2 kb around the sequences of interest are recommended Masking of DNA sequence repeats is not necessary 2 Scroll down the page and select the name of the species of interest from the Species pull down menu Fig 4 6 1 If using the sample sequence select H sapiens in that menu The user has to make sure that the Species name located below the Input Sequence section is selected correctly Choosing the right species name is essential to obtaining meaningful results as the program automatically chooses the statistical model for the sequence analysis with regard to the givenname Currently as of August 2002 models are available for H sapiens C elegans A thaliana D melanogaster C reinhardtii Z mays T aestivum H vulgare M musculus and O sativa The strand of the DNA sequence does not have to be specified because prediction is performed on both DNA strands simultaneously 3 Scroll further down the page and set the Output Options Fig 4 6 1 If using the sample sequence provided enter a valid E mail address and check all of the three check boxes Generate PostScript Graphics Print GeneMark 2 4 Predictions and Translate Predicted Genes into Proteins By default the program generates a list of predicted exons for each predicted gene The user has the option of choosing graphical output by checking Generate PostScript Graphic
168. e when Prior 0 02 its P score lt 0 5 see Figures 4 2 1 4 2 2 and 4 2 3 GUIDELINES FOR UNDERSTANDING RESULTS The result output contains the following information File_Name maybe truncated if too long Sequence_length in basepairs G C_content see Feature Variables Used in MZEF in this unit s Appendix and a table of internal coding exons predicted The nine columns in the table are Coordinates the exon coordinates in the input DNA sequence if Strand 2 one should reverse complement each output region to get the sense strand segment P the posterior probability gt 0 5 for each exon how likely is it an exon Fr1 first fame preference score how likely the 1st frame is coding Fr2 second frame preference score how likely the 2nd frame is coding Fr3 third frame preference score how likely the 3rd frame is coding Orf open reading frames e g 112 or 110 means the first and the second frames are open 3ss the acceptor site score 3 splice site score Cds the coding potential score exon coding potential 5ss the donor site score 5 splice site score In the Web example see Basic Protocol 1 Figure 4 2 1 the predicted exon in region 4076 4208 has only one ORF in the third frame which is consistent with Fr3 being relatively larger than both Frl and Fr2 For the same reason the predicted exon 7759 7856 has two ORFs in the first and the second because Orf 112 but the ORF in the f
169. ect strand and the second is in the third frame of the direct strand High coding potential predicted by GeneMark is seen for the second exon by a wide peak in the area of gt 4000 and by the thick gray line High coding potential for the first exon cannot be seen due the small size 58 nt of the exon yet it is still correctly predicted by GeneMark hmm USING UNIX VERSION OF GeneMark hmm This protocol describes application of the stand alone GeneMark hmm for prediction of genes in eukaryotic genomes An introduction to Unix is provided in APPENDIX 1C Necessary Resources Hardware Unix workstation with Linux Sun Solaris DEC Unix SGI Irix or IBM AIX operating system Software Stand alone Unix version of GeneMark hmm which is available through affiliated distributor see information located at http opal biology gatech edu GeneMark faq html Files The model for a specific organism provided as a matrix file e g human mtx The analyzed DNA sequence must be in FASTA format APPENDIX 1B The sample sequence DNA sequence of complete human serum albumin ALB gene GenBank accession no M12523 used to illustrate this protocol can be downloaded as file M12532 fna at http www3 interscience wiley com c_p cpbi_sampledatafiles htm la From the Unix command line run GeneMark hmm program name gmhmme sgmhmme lt DNA file gt m lt matrix file gt o lt output file gt The name of the compiled version of GeneMark h
170. ed at TIGR for annotation of the genomes of the parasite Theil eria parva and the fungus Aspergillus fumi gatus Other projects are planned for the future Gene modeling performance Next some estimates of the accuracy of GlimmerM on selected organisms is provided The main difficulty with training a gene finder for a newly sequenced genome is the lack of positive examples Ideally the training data set should contain genes that represent a random sample of genes in that genome but practical considerations often make this requirement im possible to satisfy The accuracy of the resulting gene finder will depend not only on the mod eling technique used but also on the training set and one can often dramatically improve gene finding results by re training a system as more genes become known Because of these aspects the numbers presented here should be Current Protocols in Bioinformatics only considered as a rough estimate of the accuracy of GlimmerM The malaria specific version of GlimmerM has an accuracy that has been measured on the two published P falciparum chromosomes Gardner et al 1998 Bowman et al 1999 using known genes from that organism in order to validate the accuracy When computed at the nucleotide level sensitivity and specificity are above 94 and 97 respectively Salzberg et al 1999 Pertea et al 2000 in other words 94 of coding nucleotides are correctly labeled as coding Using another measur
171. ed exons Xu et al 1994b There are a number of constraints placed on the gene model based on splicing considerations For example not only must an open reading frame be main tained over both exons when two predicted exons are connected but the reading frame s that have been predicted for each of the exons must be maintained These sorts of constraints can be used to evaluate the model building process at each step and to force the system to explore other alternatives if there are violations Description of GrailEXP system GrailEXP comprises three major compo nents Hyatt and Uberbacher 2002 1 Per ceval which provides exon prediction as well as CpG Island and Repetitive element predic tion 2 Galahad which provides gene mes sage alignment functionality and 3 Gawain which performs gene assembly Perceval Perceval Protein coding Exon Repetitive and CpG Island EVALuator reads in a DNA sequence and produces a list of possible GRAIL Exon Candidates It provides user options for locating repetitive elements and CpG Islands and for filtering the exon candidates against a repetitive element data base A GRAIL Exon Candidate is a region of the sequence identified by the GRAIL neural net work as being a potential exon on the forward orreverse complementary DNA sequence with a start codon or an AG acceptor splice site at its Current Protocols in Bioinformatics starting position and a stop codon or a donor spli
172. ee step 3 RUNNING GlimmerM VIA THE WEB The GlimmerM system can be run directly on genomic sequences by using the Web interface at TIGR located at http Avww tigrorg softlab glimmerm The Web server provides genefinding using GlimmerM 1 2 for three organisms P falciparum A thaliana and O sativa rice and others may be added in the future This Web interface to GlimmerM should fulfill the needs of laboratories that do not have the facilities to install and run a Unix based software system like GlimmerM and of those laboratories that might be sequencing a single BAC or some other small region of a genome The authors Web server allows anyone to submit sequences for analysis in chunks as large as 200 000 bp by uploading a FASTA formatted file APPENDIX 1B into the server Sequences lt 30 kbp can be directly pasted into the browser The user has the option of selecting which organism specific version of the gene finder is desired and also the option of whether to see the results on the screen or to have them sent by E mail see Figs 4 4 6 and 4 4 7 The performance of GlimmerM will degrade for organisms other than those used for training the system thus for anything other than organisms closely related to the three listed above re training is highly desirable as explained in the Support Protocol however the Web interface does not allow users to re train GlimmerM GUIDELINES FOR UNDERSTANDING RESULTS When annotating a genome G
173. egion is con sidered a gene For stringent gene finding high specificity a higher threshold is recom mended For a larger number of predictions high sensitivity a lower threshold should be used By default the threshold is set at 0 5 The Window Size default value is 96 The Step Size by which the sliding window moves is equal to 12 nt as a default Problems may arise if the format is not a correct one FASTA Sequences that are too large i e larger than 5 Mbp should be split into smaller sequences for analysis Sequences that are too small i e smaller than 400 bp cannot be accurately analyzed and should not be submitted Suggestions for Further Analysis The protein translations of the predicted genes can be easily used in BLASTP unrr 3 3 to obtain additional information about the pu tative protein Experimental biologists can use the sequences around the predicted genes to create primers for PCR analysis of genes of interest as well as for designing DNA expres sion arrays Literature Cited Besemer J and Borodovsky M 1999 Heuristic approach to deriving models for gene finding Nucleic Acids Res 27 3911 3920 Besemer J Lomsadze A and Borodovsky M 2001 GeneMarkS A self training method for prediction of gene starts in microbial genomes Implications for finding sequence motifs in regulatory regions Nucleic Acids Res 29 2607 2618 Borodovsky M and McIninch J 1993 GEN MARK Parallel ge
174. emarks cgi This program can analyze an anonymous sequence of a eukaryotic virus derive necessary statistical models and predict genes The major difference between this program and the prokaryotic version of GeneMarkS is that for sufficiently long genomes instead of deriving a model of the ribosomal binding site the eukaryotic version of GeneMarkS derives a Kozak like pattern near the gene start For sequences shorter than 100 kb both versions operate in essentially the same way using heuristic models As described in step 3 below these models are then used in GeneMark hmm to predict protein coding regions Necessary Resources Hardware A personal computer or workstation with Web access Software A Web browser e g Netscape Communicator or Microsoft Internet Explorer Files A single sequence in FASTA format APPENDIX 1B See Alternate Protocol 2 in UNIT 4 5 for the example sequence used in Figure 4 5 8 1 Viaa Web browser connect to http opal biology gatech edu GeneMark genemarks cgi In the Input Sequence section paste an input sequence into the Sequence box area or alternatively click on Browse next to the Sequence File Upload box to upload the input sequence file from a local drive The Sequence File Upload option is more powerful since the copy and paste method imposes a limit on the length of the sequence If the sequence has a FASTA APPENDIX 1B title line e g gt Sequence name this name will be assigne
175. embly engine therefore consuming a smaller amount of memory and running time Current Protocols in Bioinformatics Finding Genes 4 3 3 Supplement 18 Using geneid to Identify Genes 4 3 4 Supplement 18 Table 4 3 1 Information Provided by geneid for Each Coding Exon in the Gene Fields Description 1 Type of exon First Internal Terminal or Single 2 3 Location of the exon within the input sequence i e the positions of the exon defining signals 4 Score of the exon 5 Strand of the gene always the same for all the exons in the same gene 6 Frame 7 Remainder 8 9 Score of the two signals defining the exon start acceptor and donor stop 10 Score derived from the nucleotide composition of the exon sequence 11 Score derived from potential similarity of the exon sequence to known coding sequences 12 13 Location of the exon within the amino acid sequence of the predicted gene 14 Gene identifier From left to right in Figs 4 3 1 and 4 3 2 bottom t date Wed Jan 17 18 13 51 2007 t source version geneid v 1 2 geneid imin es Sequence examplel Length 32001 bps Starte predicted in sequence examplei 0 32000 Start 53 55 2 05 CTGCACCACGTGCAATGNNN Start 156 168 0 59 NNNNNNNNNNNNNNATGACG Start 178 180 2 09 TTGGCTTACCTGTAATGGGT Start 610 612 1 80 GCTGTTAAAAGCAGATGGTG Start 631 633 2 48 CTGAGGTTTGTTCAATGOCC Start 679 681 2 60 ATCTCAGG
176. ements over GRAIL by exploiting the information gleaned from sequence similarities between the sequence being analyzed and sequences in one or more databases of complete and partial known gene messages including RefSeq HTDB dbEST EGAD DOTS and RIKEN GrailEXP provides substantially more accurate gene models by making use of sequence similarity with Expressed Sequence Tags ESTs and known genes GrailEXP is also relatively unique in providing alternatively spliced constructs for each gene based on the available EST evidence GrailEXP has been designed and implemented as a modular system This facilitates its use as a pure exon finder similar to GRAIL 2 as a sequence alignment program for aligning cDNAs with genomic sequence similar to sim4 or as an expert system for constructing gene models from all available information namely predicted exons and sequence alignments The Alternate Protocol describes the use of the Genome Analysis Pipeline a Web application which allows users to perform comprehensive sequence analysis by offering a selection from a wide choice of supported gene finders other biological feature finders and database searches PERFORMING GENE PREDICTIONS USING THE GrailEXP WEB INTERFACE The GrailEXP Web interface http compbio ornl gov grailexp provides the user with selection options for organisms databases to search and analysis tasks to perform The Web interface limits input sequence size to 500 000 bases Results a
177. en that exon candi date is flagged as repetitive element and marked for elimination Next a strand resolution proc ess is applied wherein overlapping exons on Opposite strands are examined and the lower scoring cluster containing what the authors call shadow exons is marked for elimination The final list of exon candidates is then pro vided in the output with the shadow and repeti tive exons appropriately flagged A sequence can also be masked for repetitive elements prior to submission to GrailEXP s exon prediction program but this is not recommended as it may lead to the loss of legitimate exons Galahad Galahad Gene message ALign ment has three major usage modes 1 Read in a sequence and a GRAIL Exon Candidate file search a database of par tial complete gene messages and produce a list of gene message alignments 2 Read in a sequence file only search the entire sequence against a database of par tial complete gene messages and produce a list of gene message alignments Finding Genes 4 9 11 Supplement 4 GrailEXP and Genome Analysis Pipeline for Genome Annotation 4 9 12 Supplement 4 3 Read in a sequence file and an accession number of a partial complete gene message and produce an alignment of the two sequences A gene message alignment is an alignment between gene components exons in a genomic sequence and a spliced gene message It differs from a regular global alignment
178. ence To further improve the predic Finding Genes 4 6 11 Supplement 1 Eukaryotic Gene Prediction Using GeneMark hmm 4 6 12 Supplement 1 tion of the exons the models of the translation start translation end acceptor and donor were derived At the post processing stage these models are used to refine translation initiation and termination codon predictions as well as the predictions of exon intron boundaries However the range of uncertainty for the in itiation and termination codon positions still presents a problem for eukaryotic Gene Mark hmm as well as for a majority of other gene prediction programs due to relatively weak statistical patterns for these sites Model construction The goal of model construction is to maxi mize both the specificity and the sensitivity of gene predictions For construction of the model the sequences specific to a given organ ism and having a reliable sequence annotation are selected for the training set They are veri fied for sequencing irregularities and further clustered by their GC contents Each cluster must contain at least 1 2 Mb to achieve reason able accuracy The Markov models of several orders are generated for coding and noncoding regions From the same training set of se quences the position frequency matrices are constructed for translational start and termina tion sites as well as for donor and acceptor splice sites Other parameters including l
179. ence section paste the DNA sequence into the Sequence box area and provide an optional sequence title in the Sequence Title box Fig 4 6 1 If using the sample sequence see above copy and paste the human albumin DNA sequence M12523 in FASTA format into the Sequence box or upload the file M12523 fna from a local drive using the Browse button next to the Sequence File Upload box Eukaryotic Gene Prediction Using GeneMark hmm 4 6 2 Supplement 1 The Sequence File Upload option is the universal method for sending DNA sequences to the GeneMark hmm server as the copy and paste method has a limit on the length of the sequence If the sequence has a FASTA APPENDIX 1B title line e g gt Sequence name this name will be assigned to the sequence in the output unless the user has given a name in the Sequence Title area For the purpose of analysis the FASTA title line all numbers and all white space characters e g spaces tabs and line returns are ignored In addition all ambiguous letters and ASCII symbols other than the symbols of the four standard nucleotides assuming that they occur rarely are replaced with C This minimizes the chance of the possible creation of a false start or stop codon Current Protocols in Bioinformatics D Eckaryotic Genemark hem microsoft Internet Leplor er fe ght Yew Apote joos io Ep OOR DO y gated PIAA AEN 69 De Eukaryotic GeneMark AM
180. enes These consist of sequence similarity search BLASTP UNIT 3 4 and protein family classification analysis Pfam UNIT 2 5 BLASTP unrr 3 4 can be run against GenBank nr nonredundant database or SwissProt protein databases by selecting one of these options from the Database pull down menu Another menu is provided where the user can select BLASTP E value parameter threshold UNIT 3 4 Additionally several feature finders are available consisting of CpG Islands RepeatMasker tRNA BAC end pairs and STS e PCR The user can also select the option to run BLASTN unir3 3 on the input DNA sequence against one of several DNA databases Once the user has selected the analysis tools of interest and has set relevant options the sequence of interest can be loaded either by cutting and pasting the sequence into the DNA Sequence text box or by using the Browse button to select a sequence file from the local file system If the user simply wishes to perform a demo run the Demo checkbox can be checked instead of loading a sequence The request can then be submitted by clicking on the Submit Request button For the purposes of this example the same data set used in the Basic Protocol humadag will be used here On submission of the request the server returns a request status page which periodically refreshes itself to provide the user with an indication regarding the precise status of pipeline processing and failure or completion of each analysi
181. enes often include low complexity regions Extreme masking of the query sequence may lead to some genes or fraction of genes being missed G C Content Accuracy of predictions may be quite sensitive to G C content Indeed gene structure has been reported to depend on the G C content However different programs appear to behave differently with respect to G C content In general geneid predictions are poorer in low G C content sequences The Parameter File geneid needs a parameter file to build the predictions This parameter file is computed explicitly for a given species or taxonomic group Currently there are parameter files for Homo sapiens which can be safely applied to all mammalian sequences Tetraodon nigroviridis which can be safely used at least in other pufferfish species such as Fugu rubripes Drosophila melanogaster probably extensible to other diptera species Caenorhabditis elegans Dictyostelium discoideum Solanum lycopersicum Triticum aes tivum Oryza sativa Arabidopsis thaliana and many others New models are regularly uploaded at the geneid Web page The parameter file contains mostly the description of the probabilistic model on which the predictions are based see Background Information Position Weight Matrices PWM to predict sites and the Markov model to score candidate exons These need to be estimated from large training sets of sequences and users in general are not expected to modify them Howev
182. ength distribution for introns and exons are also gen erated from the training set All parameters of the model are concatenated into a single matrix file and encrypted in order to avoid uninten tional modification of crucial parameters To date GeneMark hmm models have been constructed for several eukaryotic organisms covering important classes of low and high eukaryotes such as invertebrates C elegans green algae C reinhardtii insects D melanogaster plants Z mays O sativa A thaliana T aestivum H vulgare and mam mals H sapiens M musculus Critical Parameters and Troubleshooting GeneMark hmm has been thoroughly tested since 1997 Yet there are some sequences that can cause termination of the program without output If a user encounters this problem the authors would appreciate it if the user sends the sequence and the name of the matrix used to the administrator of the GeneMark hmm Web page at the E mail link on the page Organism specific models cannot be edited by users For time efficiency of calculation values of several variables are restricted e g the maximum in tron length is set to 30 kb Some genes i e those with introns larger than this limit will not be predicted correctly since the program will be forced to predict artificial exons in the intron region larger than 30 kb Suggestions for Further Analysis The protein translations of the predicted genes can be further analyzed by
183. eorgia Institute of Technology Atlanta Georgia Alex Lomsadze Nikolai Ivanov and Ryan Mills School of Biology Georgia Institute of Technology Atlanta Georgia Current Protocols in Bioinformatics Application of FirstEF to Find Promoters and First Exons in the Human Genome Mammalian genomes contain vast amounts of cis regulatory regions responsible for differential regulation of thousands of protein coding genes Identification of these regulatory regions generally located upstream of the first exon is a very important part of gene finding First Exon Finder FirstEF Davuluri et al 2001 was developed to predict first exons and promoters in the human genome The FirstEF algorithm is a decision tree that consists of a set of quadratic discriminant functions UNIT4 2 at its nodes The discriminant functions are optimized to find potential first donor sites and CpG re lated and non CpG related promoter regions For every potential first donor site GT and upstream promoter region FirstEF decides whether or not the intermediate region can be a potential first exon based on a set of quadratic discriminant functions An explanation of both the Web based see Basic Protocol and local see Alternate Protocol versions of FirstEF to find potential promoters and first exons in human DNA are given In addition a discussion of how the user can combine the predictions of FirstEF with other information such as mMRNA EST alignments or predic
184. er to reduce computation time and memory required geneid uses a number of cutoffs to further consider predicted sites and exons In some cases users may want to modify these cutoffs to increase or decrease the size of the set of candidate exons and sites For instance users may want to predict and score every GT dinucleotide as a candidate donor site In such a case the cutoff associated with the PWM for donor sites should be set to a very low number 99 for instance See the geneid manual for details For same species parameters are specifically estimated for regions with different G C content isochores The Gene Model From a large number of candidate exons geneid selects a proper combination of exons to assemble the predicted gene structure This assembly must conform to a number of biological constraints for example that selected exons cannot overlap or that an Open Reading Frame ORF should be maintained along the assembled gene These biological constraints are defined in a set of rules in the so called gene model included within the parameter file These rules refer to the order of gene features in the prediction and to the distances between them Each rule is a three column record in the gene model For instance the rule First Internal Internal Terminal 40 10000 indicates that elements exons of type Internal in the forward strand and of type Terminal in the forward strand are allowed only immediately after exons o
185. er has no size limit Submit target sequence to N SCAN 1 Register on the N SCAN page at Attp mblab wustl edu nscan register Fill in your preferred username your email address your first and last name and your institutional affiliation Press the Create account button at the bottom of the green box An account is created for you and all your predictions will be kept on the N SCAN server until you delete them A new page appears telling you that your password is being emailed to you Check your email to retrieve the password By default gene prediction results are accessible on the N SCAN output page from which they can be downloaded If you want to have the results emailed to you after logging in go to Preferences and check the appropriate box Current Protocols in Bioinformatics N SCAN Twinscan Gene Predictor Copy and paste gene prediction 1 2 2 Sequence Masking Organism Information You can either upload a text file or cut Your sequence will be masked for Select the organism that your sequence and paste your sequence into the box interspersed repeats but not for low came from below complexity you want to change this m in i check the boxes below I your sequence se Text is lowercase masked and you want to Ciade use this information check the last box TGACCGCCGGCGGCCTITAGATTT TCCATGGCCTAGATTTGTGAAATATC Mask Interspersed repeats S human IGTTGACGAGATACCCCTTATATGTATC IGGTTACCATAGTCGTGTGC
186. er than the sequence file itself Strand I or 2 One should try both strands if the coding strand information is unknown Current Protocols in Bioinformatics Prediction Results Sequence gt GI 178343 GB M12523 1 HUMALBGC HUMAN SERUM ALBUMIN GENE COMPLETE CDS Length 19002 bp C G Content 35 Type End5 End3 Leng Fr St Ac Do Te FrCod Prob Intr 1817 1854 38 2 0 561 0 689 0 656 0 992 Intr 2564 2621 58 0 0 443 0 567 0 155 0 635 Intr 4076 4208 133 2 0 536 0 587 0 567 0 993 Intr 6041 6252 212 1 0 538 0 646 0 691 0 971 Intr 6802 6934 133 1 0 547 0 545 0 553 0 932 Intr 7759 7856 98 0 0 536 0 607 0 965 0 999 Intr 9444 9573 130 0 0 574 0 553 0 636 0 999 Intr 10867 11081 215 0 0 541 0 597 0 991 0 997 Intr 12481 12613 133 1 0 576 0 548 0 617 0 999 Intr 13702 13799 98 0 0 548 0 719 0 976 1 000 Intr 14977 T5115 139 1 0 526 0 457 0 591 0 864 Intr 15534 15757 224 2 0 462 0 562 0 724 0 678 Intr 16941 17073 133 0 0 483 0 609 0 667 0 999 Reverse Strand Notations Star initial exon Intr internal exon Term terminal exon End5 5 exon coordinate End3 3 exon coordinate Leng exon length Fr frame number 0 1 or 2 score Coding cDNA and protein for each coding region ALB Score H Pepe f f BewowpokR BUOUYNOFRO St Ac start or acceptor site score Do Te donor or stop site score FrCod in frame coding score Prob exon probability Score lt 02 41 LS 38 ol
187. er to obtain the system a representative of a nonprofit organization should fill out a license agreement available on the TIGR Web site http www tigr org under Software Interested commercial organizations should see the Web site for additional instructions For nonprofit organiza tions the system is made available almost immediately after submitting the license agreement Files A FASTA file APPENDIX 1B containing the sequence to be analyzed There is no maximum sequence length set by default in the program The FASTA file used in the example below is available at the Current Protocols Web site hittp www3 interscience wiley com c_p cpbi_sampledatafiles htm Install software 1 Submit a license agreement and download install and compile the software To install and compile the program type gt tar xvfz GlimmerM tar gz gt cd GlimmerM sources gt make 2 If necessary train GlimmerM for a new organism see Support Protocol Run GlimmerM to analyze DNA sequences for their coding potential 3 The program GlimmerM takes two inputs a DNA sequence file in FASTA format APPENDIX 1B and a directory containing the training files for the program If not specified the training directory is assumed to be the current working directory For Current Protocols in Bioinformatics instance if the user is running a pre compiled version of GlimmerM located in the bin directory the following command should be used glimmerm_ lt
188. es large selection of repetitive element libraries which are required for running RepeatMasker The library is free for download by academic users who are required to set up accounts to access the database files by filling an online form http www girinst org accountservices register php Commercial users should contact Jolanta Walichiewicz jola girinst org Once again if one s genome of interest does not have an appropriate repeat library file in Repbase Update one can establish one with RECON Bao and Eddy 2002 or RepeatScout http bix ucsd edu repeatscout Price et al 2005 Stein et al 2003 used RECON to establish a repeat library file for the round worms C elegans and C briggsae RECON can also be obtained as part of a Repeat Modeler package available for download from http www repeatmasker org RepeatModeler html Alternatively the RepeatScout software can also be used with RepeatMasker to identify and mask repeat family sequences from newly sequenced genomes A FASTA file APPENDIX 1B or a collection of FASTA files can be processed via the command line RepeatMasker Note that there is essentially no size limit for query sequences for running RepeatMasker on the command line The example file used in this protocol is the fully sequenced whole Caenorhabditis elegans genome 102 287 094 bp in length downloaded from the WormBase http www wormbase org FTP site ftp ftp wormbase org pub wormbase genomes elegans
189. es assuming that they occur rarely are replaced with C This minimizes the chance of the possible creation of a false start or stop codon 2 Scroll further down the page and set the Output Options The user may request the graphical output An E mail address is required for sending text output for sequences longer than 1 Mbp or if graphical output is requested 3 After completing the above entries click the Start GeneMarkS button The results will be depicted on the browser or will be sent to the E mail address provided 4 Interpret the text output The text output from GeneMarkS is identical to that of GeneMark hmm see Alternate Protocol 1 5 Interpret the graphical output The graphical output from GeneMarks is identical to that of GeneMark hmm see Alternate Protocol 1 when using just one model Current Protocols in Bioinformatics Finding Genes 4 5 13 Supplement 1 Prokaryotic Gene Prediction Using GeneMark and GeneMark hmm 4 5 14 Supplement 1 GUIDELINES FOR UNDERSTANDING RESULTS The final steps of each protocol within this unit offers some guidance as to how to evaluate the results The output of the programs has a clear cut meaning i e the parsing of the DNA sequence into predicted coding and noncoding regions Questions about the reliability of individual gene predictions have been addressed in previous publications Borodovsky and McIn inch 1993 Lukashin and Borodovsky 1998 Bese
190. es to the probabilities that are computed by the quadratic discriminant functions in the FirstEF algorithm Exon QDF Promoter QDE and Donor QDF described in Davuluri et al 2001 By default FirstEF prints out the predictions of all first exons that satisfy the three constraints P exon gt 0 5 P donor gt 0 4 and P promoter gt 0 4 This choice of cut off values 0 5 0 4 0 4 results in a sensitivity and specificity of 80 based on cross validation analysis of real first exons Davuluri et al 2001 Current Protocols in Bioinformatics Advanced users may wish to run FirstEF with different cut off values in order to obtain more sensitive or more specific first exon predictions In principle the user can adjust all three cut off values independently of each other which might be of interest in special circumstances where for example first exons with strong splice donor sites i e P donor 2 0 8 and weak promoters i e P promoter lt 0 5 or weak splice donor sites i e P donor lt 0 5 and strong promoters i e P promoter 0 8 are searched As a rule of thumb it is recommend to modify only one parameter e g P exon and choose the other two cut off values proportional to this value e g P donor 0 8 X P exon and P promoter 0 8 x P exon Note that cut off values below 0 2 are not accepted A value of 0 8 was chosen to maintain balance between the parameters and has no special significance in and of
191. escribe it here because manual tuning sometimes can improve the accuracy of the gene predictions The flags on lines 6 and 7 inthe config file see Table 4 4 2 are just internally used by the system They signal if decision trees were created from the available data or not GlimmerM will use decision trees in computing the splice site scores only if these flags were created by the training procedure see Background Information for a brief description of how the splice sites are determined A decision tree is a supervised learning method that learns to classify objects from a set of examples It takes the form of a tree structure of nodes and edges in which the root and the internal nodes test on one of the objects te Finding Genes 4 4 7 Current Protocols in Bioinformatics 5 Diasas l Zpredator apertea Halaria train trainti_O1 T rainGlimH200i 09M hore config_file 0 m predator apertea Halaria train traini1_OL no Training data created successfully Check e for trair ning s for training 6080 value for the acceptor sites without filterin value for the donor sites without filtering value for the s with Filtering value for with filtering aining 175 or training Default threshold value for the Decision trees created successfully IMMs trained s fully Fully All training files are in d false nofilter when using or not using filters respectively ferent threshold that can be
192. exact gene level As a result specificity and sensitivity on the gene level tend to be positively correlated In general gene predictors are most accu rate at the nucleotide level less accurate at the exact exon level and least accurate at the ex act gene level At the exon level initial and terminal exons are predicted with less accu racy than internal exons in humans N SSCAN Current Protocols in Bioinformatics predicts 59 of initial exons and 66 of termi nal exons exactly right versus 89 of internal exons This is one of the reasons that gene pre dictors can identify most coding nucleotides and even most coding exons but finding cor rect gene structures is still a challenge This is especially true in larger genomes in which less of the sequence is coding As a result gene predictors tend to differ from one another least in their nucleotide accuracy more in their exon accuracy and most of all in their exact gene accuracy Accuracy on specific genomes For many genomes N SCAN is currently the best available de novo gene predictor see below However its accuracy is dependent on optimization for every genome for which it is used The Brent laboratory recently created a free software tool that automates this mod eling procedure called iParameterEstimation http mblab wustl edulsoftware iparameter estimation In this section brief comments and impressions are presented with regard to each of the genomes fo
193. examines a 72 base region around the sequence using a simple Markov model and reports back a score for that site If a poly A site is within 5000 bases of the stop codon of a gene and does not fall in an intron then it is retained At most one poly A site the highest scoring in the case of multi ples is assigned to each gene model The pro moter recognition system looks for a TATA or ATA and examines the region around these bases In particular the neural net is fed infor mation on GC content information on CAAT GGGCGG and ATG patterns located in close proximity and sequence fragments flanking these patterns The neural net evaluates the scores and assigns a total score to the promoter Again the promoter must be within 5000 bases of a start codon of a predicted GrailEXP gene model to be retained and cannot fall within an intron At most one promoter element is as signed to each gene model Alternate methods for accessing GrailEXP The Genome Channel Genome Channel http compbio ornl gov channel is a Web based annotation system comprising GrailEXP gene predictions for a variety of organisms including human mouse and many microbial Finding Genes 4 9 13 Supplement 4 GrailEXP and Genome Analysis Pipeline for Genome Annotation 4 9 14 Supplement 4 genomes Genome Channel is the recom mended starting point for users interested in viewing precomputed gene models A descrip tion of Genome
194. f form generated by the bioinformatics group at UCSC are available for download at http hgdownload cse ucsc edu downloads html Finding Genes 4 8 9 Current Protocols in Bioinformatics Supplement 20 ALTERNATE PROTOCOL 2 Using N SCAN or TWINSCAN 4 8 10 Supplement 20 3 Run the N SCAN executable by entering the following command on a single line nscan o parameter file masked target sequence a target align gt output file The output file is in GTF format It is possible to run the N SCAN code in TWINSCAN mode by omitting the a option using the c option with a conservation sequence file and using a TWINSCAN specific parameter file For generating a conservation sequence and more details on TWINSCAN see the README provided with the N SCAN package Using Nscan_driver p1 on a Local Computer The three steps described in Alternate Protocol 1 can be run automatically using a Perl script called Nscan_driver p1 which is included in the bin directory of the N SCAN distribution However this script may not satisfy the needs of all users as it does not give access to all options available in the N SCAN code The script needs a configuration file that lists the full paths to all programs run by Nscan_driver pl Users will need to customize this file for their particular applications It may be necessary to adapt Nscan_driver p1l to the specific environment of the user s system but this advanced pro
195. f type First in the forward strand or of type Internal in the forward strand The third column indicates the valid Current Protocols in Bioinformatics Gene_Model intronic connections First Internal Internal Terminal 20 40000 block Terminal Internal First Internal 20 40000 blockr connections to regulatory elements Promoter First Singlet 50 4000 Terminal Single aataaat 50 4000 First Single Promoter 50 4000 aataaa Single Terminal 50 4000 intergenic connections aataaat Terminal Singlet Single First Promoter 500 Infinity aataaat Terminal Singlet Single Terminal aataaa 500 Infinity Promoter First Single Single First Promoter 500 Infinity Promoter First Single Single Terminal aataaa 500 Infinity Figure 4 3 14 geneid Default Gene Model distances at which these elements can be assembled into a predicted gene In this case these elements must be at least 40 bp and at most 10 000 bp apart Note that this rule specifies the constraints governing intronic connections in the forward strand The basic gene model distributed with geneid v1 2 appears in Figure 4 3 14 Note that the default gene model includes rules for promoter elements and poly A signals The current version of geneid however predicts only elements of type First Internal Terminal or Single Predicted promoter elements or poly A signals probably obtained using other programs must be passed as external information vi
196. factors In general prediction is more accurate in compact genomes like those of C elegans and A thaliana than in big genomes with long introns and large numbers of pseudo genes like those of mammals For mammalian sequences accuracy can be improved by au tomatically removing processed pseudogenes from the predictions using PPFINDER van Baren and Brent 2006 The benefit of N SCAN s alignment se quence method depends in part on the evolu tionary divergence between the target and in formant genomes This distance must be large enough so that noncoding regions are less con served than coding regions but small enough to find most coding sequences in both species Based on preliminary data the authors believe that the method works best for an informant that maximizes the sum of the mismatch and gap percentages in the whole genome align ment of target and informant The patterns of mismatches and gaps helps to discrimi nate between coding and non coding regions Finding Genes 4 8 13 Supplement 20 Using N SCAN or TWINSCAN 4 8 14 Supplement 20 e g most mismatches in the coding region oc cur in the third codon position For a closely related species the number of mismatches and gaps is small and therefore not very informa tive As the evolutionary distance between tar get and informant increases the number of mismatches and gaps increases and so does informant utility until it reaches a maximum
197. false positives An easier to understand measure that combines the sensitivity and specificity values is called the correlation coefficient Like all correlation coefficients its value can range from 1 meaning that the prediction is always wrong through zero to 1 meaning that the prediction is always right As a result of a Cold Spring Harbor Laboratory meeting on gene prediction Finding Genes Computational Analysis of DNA Sequences March 1997 a Web site called the Banbury Cross was created The intent behind its creation was twofold to allow groups actively involved in program development to post their methods for public use and to allow researchers actively deriving fully characterized finished genomic sequence to submit such data for use as benchmark sequences In this way the meeting participants created an active forum for dissemination of the most recent findings in the field of gene identification Using these and other published studies Jean Michel Claverie at CNRS in Marseilles compared the sensitivity and specificity of fourteen different gene identifica tion programs Claverie 1997a and references therein including all of those discussed here except PROCRUSTES PROCRUSTES was not considered because its method is substantially different from those of other gene prediction programs In examining data from these disparate sources either the best performance found in an independent study or the worst perf
198. fferent stages in the maturation of sequence data However this should not be interpreted as a blanket recommendation to use only these two programs in gene identification if it were the editors would not be presenting the contents of this chapter for consideration Remember that these results represent a compilation of findings from different sources so keep in mind that the reported results may not have been derived from the same data set It has already been stated numerous times that any given program can behave better or worse depending on the input sequences It has also been demonstrated that the actual performance of these methods is highly sensitive to G C content with no pattern emerging across all of the methods as to whether a method s predictive powers improve or degrade as G C content is raised For example Snyder and Stormo 1997 reported that GeneParser Snyder and Stormo 1993 and GRAIL2 with assembly UNIT 4 8 performed best on test sets having high G C content as assessed by their respective CC values while geneid Guig6 et al 1992 UNIT4 9 performed best on test sets having low G C content Interestingly both GENSCAN and HMMgene were seen to perform steadily regardless of G C content in the Rogic study Rogic et al 2001 This is an important result given that gene dense regions tend to be G C rich while gene poor regions tend to be A T rich International Human Genome Sequencing Consortium 2001 As al
199. g J Comp Biol 5 681 702 Guig6 R Knudsen S Drake N and Smith T 1992 Prediction of gene structure J Mol Biol 226 141 157 Guig6 R Flicek P Abril J F Reymond A La garde J Denoeud F Antonarakis S Ash burner M Bajic V B Birney E Castelo R Eyras E Ucla C Gingeras T R Harrow J Hubbard T Lewis S E and Reese M G 2006 EGASP The human ENCODE Genome Anno tation Assessment Project Genome Biol 7 S2 1 3 31 Hinrichs A S Karolchik D Baertsch R Bar ber G P Bejerano G Clawson H Diekhans M Furey T S Harte R A Hsu F Hillman Jackson J Kuhn R M Pedersen J S Pohl A Raney B J Rosenbloom K R Siepel A Smith K E Sugnet C W Sultan Qurraie A Thomas D J Trumbower H Weber R J Weirauch M Zweig A S Haussler D and Kent W J 2006 The UCSC Genome Browser Database Update 2006 Nucl Acids Res 34 D590 D598 International Chicken Genome Sequencing Consor tium 2004 Sequence and comparative analysis of the chicken genome provide unique perspec tives on vertebrate evolution Nature 432 695 716 Jaillon O Aury J M Brunet F Petit J L Stange Thomann N Mauceli E Bouneau L Fischer C Ozouf Costaz C Bernot A Nicaud S Jaffe D Fisher S Lutfalla G Dossat C Segurens B Dasilva C Salanoubat M Levy M Boudet N Castel lano S Anthouard V Jubin C Castel
200. generally known as CpG islands Gardiner Garden and Frommer 1987 As many human Current Protocols in Bioinformatics promoters are near CpG islands Davuluri et al 2001 classified first exons as CpG related and non CpG related based on CpG score This helped to better characterize the differences in sequence composition between first exons and other regions of the genome CpG score is defined as the maximum of CpG percentages of all possible sliding windows of length 201 bp within the region of 500 of the transcription start site to 500 of the first donor site The sequence window that gets the maximum CpG percentage is defined as CpG window A first exon is CpG related if there exists a CpG window of size 201 with CpG percentage 6 5 It was estimated that 70 of the first exons in the human genome are CpG related FirstEF uses different quadratic discriminant functions to identify CpG related and non CpG related first exons promoters and first donor sites Refer to unir 4 2 for a brief description on dis criminant analysis How the algorithm works FirstEF scans the input DNA for potential first donor sites During the first step the pro gram pauses at every GT and computes the a posteriori probability of the donor site P do nor by a quadratic discriminant function do nor QDF which was trained on donor sites of first exons If P donor donor cut off value FirstEF considers this as a candidate donor site other
201. genomic sequences eliminate knock out those exons occurring between the two EST3 matches and run geneid from the remaining set of exons Finding Genes 4 3 13 Current Protocols in Bioinformatics Supplement 18 Using geneid to Identify Genes 4 3 14 Supplement 18 7 10 Predict all exons on sequence example3 example3 fa geneid P param human3iso param xoGP samples example3 fa gt example3 exons gff Option x instructs geneid to print all exons option o forces geneid to switch off gene prediction and option G produces GFF output Open the fileexample3 exons gff witha text editor First discard all predicted exons between the two exons supported by EST3 i e those in the range from the position 19 031 to the position 30 233 Then open example3 EST3 gff and add the content at the end of the file Save the new file as example3 filtered exons gff and close the editor Finally use the Unix command sort on this file to obtain the ordered list of exons type sort 3n example3 filtered exons gff This operation can be also accomplished using a number of Unix file editing tools such as awk For instance the awk command would be awk 5 lt 19031 4 gt 30233 example3 exons gff cat samples example3 EST3 gff sort 3n gt example3 filtered exons gff The coordinates of the known EST have to be included in the file of candidate exons because in geneid v1 2 the R and O opt
202. get the paths to the directories To configure the program use the following script mta57 grouse repeat cd RepeatMasker mta57 grouse RepeatMasker perl configure Enter the required paths for example to write the path to the Perl interpreter enter Enter path usr bin perl To write the path to the location where the RepeatMasker program has been installed enter Enter path home mta57 repeat RepeatMasker For the path to the location where the TRF program can be found enter Enter path home mta57 repeat To add a search engine enter Enter path home mta57 repeat cross_match 4 Place repeat libraries in the correct directory i e the same directory as the script RepeatMasker Make sure that subdirectory Libraries in the RepeatMasker directory contains RepeatMasker 1liband RepeatMaskerLib emb1 files 5 Create a new directory for input and output files Note that RepeatMasker output files will be written to the same directory as the input file resides For this example type the following mta57 grouse repeat mkdir RepeatMasker_file mta57 grouse repeat cd RepeatMasker_file mta57 grouse RepeatMasker_file s Next download or copy the FASTA file current dna fa gz containing the sequence of C elegans genome to the directory and unpack it mta57 grouse RepeatMasker_file gunzip current dna fa gz Current Protocols in Bioinformatics 6 To get a brief description of the command line parameters and
203. gn ing a cDNA sequence with a genomic DNA sequence Genome Res 8 967 974 Fraser C M Casjens S Huang W M Sutton G G Clayton R Lathigra R White O Ketchum K A Dodson R Hickey E K Gwinn M Dougherty B Tomb J F Fleischmann R D Richardson D Peterson J Kerlavage A R Quackenbush J Salzberg S Hanson M van Vugt R Palmer N Adams M D Gocayne J Venter J C et al 1997 Genomic sequence of a Lyme disease spiro chaete Borrelia burgdorferi Nature 390 580 586 Fraser C M Norris S J Weinstock G M White O Sutton G G Dodson R Gwinn M Hickey E K Clayton R Ketchum K A Sodergren E Hardham J M McLeod M P Salzberg S Peterson J Khalak H Richard son D Howell J K Chidambaram M Utter back T McDonald L Artiach P Bowman C Cotton M D Venter J C et al 1998 Complete genome sequence of Treponema pallidum the syphilis spirochete Science 281 375 388 Gardner M J Tettelin H Carucci D J Cum mings L M Aravind L Koonin E V Shal lom S Mason T Yu K Fujii C Pederson J Shen K Jing J Aston C Lai Z Schwartz D C Pertea M Salzberg S Zhou L Sutton G G Clayton R White O Smith H O Fraser C M Hoffman S L et al 1998 Chro mosome 2 sequence of the human malaria para site Plasmodium falciparum Science 282 1126 1132 Finding Genes 4 4 19 Us
204. gure 4 8 3 An example of the top portion of the Submission Results Web page that automati cally replaces the waiting page when N SCAN has completed processing of a sequence submitted to the server that N SCAN was run as expected and for reproducing results Changes to any of these specifics can change the results 8 Look at the schematic summary of predicted genes The top of the page Fig 4 8 3 displays the number of genes predicted by N SCAN and a graphical representation of the predicted genes on the input sequence A thin line represents the input sequence with the length shown on the right side and exons are shown as colored boxes below this line Exons with the same color are part of the same gene A short summary of each gene is listed below the picture Current Protocols in Bioinformatics Finding Genes 4 8 5 Supplement 20 Using N SCAN or TWINSCAN 4 8 6 Supplement 20 Gene Details Gene 1 Exon Type Strand Begin End Length 1 CDS UTR 451 1436 985 MQKAIKKELSFSLDTLERYRAKYGRSASLDTNGTPIAHTDGDQAAPAPPP PSIFTKSLSPPKMTKLOELQQKKEAYLRAKEHEREMEQLORTERRSIING SDTPKPKTGSPTSTSPTPNASSSSTAAVATKTSSPAGYTNWSNHHATLCS QSWVAISELMYFCDKYEFTTLSTRDLRTHQEIVAEVRALLSGKAPFDORT RFPGNIHDPENLWVCIGRCASVEYHLORIISIFRKPLNQLTPDKORTVRQ NFHLAVSELRLDISARISEVRLYDRLVFEREFRLEWLDEEA Show Transcript Gene 2 Exon Type Begin End 1 cbs 2293 2462 CDS 2519 2677 cbs 2751 4158 cos 4250 4330 cbs 4390 4774 CDSs 4841 506
205. haliana Library files for organisms that do not have Repbase Update library files can be generated ab ini tio using RECON Bao and Eddy 2002 http selab janelia org recon html or Re peatScout http bix ucsd edu repeatscout Price et al 2005 The newest version of RECON v 1 06 was released recently and is available from the RepeatModeler package at http www repeatmasker org RepeatModeler html Sequence comparisons in Repeat Masker are usually carried out by the program cross_match developed by Phil Green http www phrap org consed consed html howToGet One can also use WU BLAST http info cchmc org help wublast html see Alternate Protocol to replace cross_match for fast processing Current Protocols in Bioinformatics 4 10 1 4 10 14 March 2009 Published online March 2009 in Wiley Interscience www interscience wiley com DOI 10 1002 0471250953 bi0410s25 Copyright 2009 John Wiley amp Sons Inc UNIT 4 10 Finding Genes 4 10 1 Supplement 25 BASIC PROTOCOL 1 USING RepeatMasker VIA THE WEB INTERFACE RepeatMasker may be accessed through the Web at http www repeatmasker org cgi bin WEBRepeatMasker Unlike the command line version of RepeatMasker see Basic Protocol 2 Web RepeatMasker has a nucleotide sequence size limit of 100 kb The attempt to analyze a sequence larger than 100 kb fails whereupon a prompt is displayed in a message window shown in Fig 4 10 1 Sequences shorter than 100
206. he file containing the DNA sequence in FASTA format maximum size 100 kb or cut and paste the sequence into the sequence window A screen shot of the FirstEF Web page is given in Figure 4 7 1 Contributed by Ramana V Davuluri Current Protocols in Bioinformatics 2003 4 7 1 4 7 10 Copyright 2003 by John Wiley amp Sons Inc UNIT 4 7 BASIC PROTOCOL Finding Genes 4 7 1 Supplement 1 Application of FirstEF to Find Promoters and First Exons in the Human Genome 4 7 2 Supplement 1 tt me nt ote ont tee pomp am te ene OA arent intent ghee pondet hp FW ast in ot ver o tet toe P7 bhae s DDD Qla jra Gee J y JE f ITTI A La REEE Jid Ben FirstEF first exon and promoter prediction program for human DNA REVAD T h den siem ped som d hee bame m creer Reread Ly emie mire terme amd Here ek Le ome Agp eemest fane Nore the mimea gt oeptsdie veqmem file eagrb n DOO AD Meese parte yem sequen m FASTA foma Tite iaei seod te baga wd Ge nmd gt cut and paste the DNA 3 sequence to be analyzed in this sequence window 3 OR Pr errr came oo seems So epre et BTANA Some input the file name to upload the DNA sequence tell vedi fan the fins earn poten meh aber F eel von fn tee amet rete meh ei F change the probability cut off etal valno fan he saiet dno pmm skie Fo values according to get low and high scoring predictions Ol oaf waters wnat te 0 2 Piense mad th
207. high enough geneid exons overlapping the region will likely be included in the final gene prediction 2 Comparative gene prediction The au thors have developed the program SGP2 Parra et al 2003 which combines TBLASTX Altschul et al 1990 UNIT 3 4 and geneid to use information from sequence similarity between genomes of two different species in gene predictions for a review on compara tive gene prediction see Brent and Guig 2004 The SGP2 tool has been a com ponent of the comparative gene prediction pipelines to annotate genes simultaneously in the human mouse rat and chicken genomes Mouse Genome Sequencing Consortium 2002 Rat Genome Sequencing Project Consortium 2004 International Chicken Genome Sequencing Consortium 2004 3 Prediction of selenoproteins In se lenoproteins incorporation of the amino acid selenocysteine is specified by the UGA codon usually a stop signal The alternative decoding of UGA is conferred by an mRNA structure the SECIS element located in the 3 untranslated region of the selenoprotein mRNA Because of the nonstandard use of the UGA codon current computational gene prediction methods are unable to identify selenoproteins in the sequence of the eukary otic genomes The authors have developed a version of geneid which is able to predict genes with exons containing TGA stop codons in frame Through the option R SECIS predictions obtained by some other prediction program such
208. higher its likelihood Note however that in geneid the score of an exon depends directly on its length and that a very short exon cannot by definition have a high score Thus very short exons may have very low even negative scores UTRs geneid as with most genefinders predicts only the coding fraction of a gene Usually users are interested mainly in the gene protein product and this is not an important limitation However untranslated exons may contain good splice signals and although their nucleotide composition does not reflect the codon bias characteristic of protein coding regions they appear to exhibit a higher nonrandom bias than intronic or intergenic DNA It is thus possible that in some cases geneid predictions may include portions of a gene UTR Masking the Sequence Some types of interspersed repeats and low complexity regions exhibit a highly nonran dom sequence composition often similar to that characterizing protein coding regions Current Protocols in Bioinformatics Finding Genes 4 3 21 Supplement 18 Using geneid to Identify Genes 4 3 22 Supplement 18 Stormo 2000 geneid may include these in the gene predictions It may be advisable thus to mask the query sequence for such repeats and regions using for instance the program RepeatMasker http repeatmasker org before running geneid This strategy may increase the specificity of the predictions Let us note however that real g
209. http www3 interscience wiley com c_p cpbi_sampledatafiles htm This gene has an alternative last exon the CDS annotation is as follows CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 18526 18555 CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 17688 17732 that may be compared with the MZEF predictions below The FORTRAN program also requires the following data files which are available from the FTP site see steps 1 and 2 below asl dat as2 dat bri1 dat br2 dat ds1 dat ds2 dat h6ex1 dat h6ex2 dat h6exc1 dat h6exc2 dat h6exil dat h6exi2 dat h6 exll1l dat h6 ex12 dat h6exr1 dat h6exr2 dat qda dat and test dat is just a short input DNA sequence for a test run NOTE The names of the data files for each organism are the same but the contents of the files differ 1 Create a new directory to hold the MZEF files and change to that directory mkdir MZEF cd MZEF A copy of the FASTA file for the DNA sequence of interest e g m12523 fasta see Necessary Resources must be copied into the MZEF directory For information on navi gating through a Unix environment see APPENDIX IC If you intend to download the program and its associated data files for more than
210. humanO mtx Homo sapiens Exon Length 38 58 133 212 133 98 130 215 133 98 139 224 133 Start End Frame U m U m Wer Ore We WN WNWNWNUON WNW Ne Figure 4 6 8 The text output for the Unix version of Eukaryotic GeneMark hmm program Current Protocols in Bioinformatics Finding Genes 4 6 9 Supplement 1 ALTERNATE PROTOCOL 2 Eukaryotic Gene Prediction Using GeneMark hmm 4 6 10 Supplement 1 2 Interpret the results As in the Web interface version the first lines of the output file Fig 4 6 8 contain a description of the parameters for GeneMark hmm such as version sequence file name sequence length GC content matrix file name and time and date of prediction Following this information there is a table of predicted genes and exons in the format shown in Figure 4 6 2 see Basic Protocol step 5 for detailed explanation The predicted protein sequences in FASTA format are listed in the end of the output file As expected the output file for Unix version of GeneMark hmm contains the same predictions for complete human serum albumin ALB gene as the Web interface version see Basic Protocol USING GeneMarkS FOR GENE FINDING IN EUKARYOTIC VIRUSES This protocol is used for analyzing eukaryotic viruses with a modified version of GeneMarkS described in Alternate Protocol 2 of unir 4 5 Fig 4 5 8 GeneMarkS can be accessed through the Web at itp opal biology gatech edu GeneMark gen
211. i P Kohany O and Walichiewicz J 2005 Rep base update a database of eukaryotic repeti tive elements Cytogenet Genome Res 110 462 467 Price A L Jones N C and Pevzner P A 2005 De novo identification of repeat families in large genomes Bioinformatics 21 Suppl 1 1351 358 Smith T F and Waterman M S 1981 Identi fication of common molecular subsequences J Mol Biol 147 195 197 Stein L D Bao Z Blasiar D Blumenthal T Brent M Chen N Chinwalla A Clarke L Clee C Coghlan A Coulson A D Eustachio P Fitch D H A Fulton L Fulton R Griffiths Jones S Harris T W Hillier L W Kamath R Kuwabara P E Marra M Mardis E Miner T Minx P Mullikin J C Plumb R W Rogers J Schein J Sohrmann M Spieth J Stajich J E Wei C Willey D Wilson R Durbin R and Waterston R 2003 The genome sequence of Caenorhabditis briggsae A platform for com parative genomics PLoS Biol 1 E45 Internet Resources http www repeatmasker org RepeatMasker Web server http www girinst org Repbase Update http selab janelia org recon html RECON Web site http bix ucsd edu repeatscout RepeatScout Web site http www phrap org consed consed html howToGet cross_match Web site http blast wustl edu WU BLAST Web sites http genome ucsc edu cgi bin hgGateway UCSC Genome Browser ftp ftp wormbase org pub wormbase genomes eleg
212. id_output ps off2ps is a Unix command line program which reads a GFF file It produces PostScript output which can be redirected to a file The contents of the file can be displayed by means of programs such as ghostview or xpsview or they can be directly sent to a PostScript printing device Figure 4 3 5 shows the default gff2ps output for the prediction obtained in Basic Proto coll step 6 The plot is fitted into a single block assuming the length of the sequence to be the end of the most downstream feature which is printed so as to fit into a single physical page Genes predicted on the forward strand are displayed above the central bar and genes predicted in the reverse strand are displayed below Exons are plotted with a height proportional to their score using a three color code schema The color of the upstream half in the exon denotes the exon frame and the color of the downstream half the remainder Nonoverlapping exons are frame compatible if the remainder of the upstream exon matches the frame of the downstream one gff2ps output can be highly cus tomized Users are therefore encouraged to develop their own configuration files to suit their specific needs In particular gff2ps can also be used to plot exhaustive predictions of potential sites and exons along the query sequence In such a case users are advised to process the geneid GFF output file and use the feature field as the source see gff2ps user manual for details Cur
213. ienn ORNL Genome Analysis Pipeline Results for D 105 _ Edit View Go Comrmunicator Help y Bookmart Yr Dia http compbio ornl qow 0P3 cqi bin IF en What s Related og D iniii D Computational Biology at ORNL Home About us WINIE OTTE Projects amp Research at Channel Generation e Grail e GrailExP Pipeline Parser PROSPECT ORNL Genome Analysis Pipeline Results far sr QueryID 1059421518 28568 Object Grall EXP Ge Genelast Gras l DP Gene Var Str Exons Begin ind Start Stop Evidence 1D iat Coden Codon i End 1 1 g i 36040 36422 3605S 36147 36040 36422 4 1 ft 12 3996 35979 4031 35664 3806 35979 2 2 f 11 3896 35979 4031 35664 3896 35979 2 3 e 1 3896 35979 40n 35644 3896 35979 GeneStructures rail D Geneln 1jil Structure gt Strand r Exons 1 Begin 35040 End 36422 Elenent _ Element Pa Len ser Type Boyin ind exon 26040 36422 0 383 0 99 GeneID 211 Structure gt Strand f Exons 12 Degin 396 End 35979 Element Element Fra Len Ser Begin End promoter 1978 2055 78 0 65 exon 3096 4063 i 168 1 00 exon 19230 19291 2 62 1 00 exon 26344 26466 i 123 1 00 exon 28908 29051 0 44 1 00 exon 29823 29938 0 116 1 00 exon 31176 31703 3 28 3 00 exon 32425 32496 0 72 1 00 exon 32573 2674 i 102 1 00 exon 72051 32915 Q 6S 4 00 exon 34354 34483 i 130 1 00 enon 35100 2 2 103 1 00 exon 35651 r 0 329 1 00 polyA 35947 35952 0 6 00 GeneID 2 2 Structure gt Strand f Exons 11 Dagin 3096 Dd 35979
214. ies are expected Suggestions for Further Analysis GlimmerMExon Sometimes the DNA sequences that are be ing analyzed are nothing more than very short fragments but scientists are still interested in finding any gene fragments on these sequences If the input sequence contains only partial gene models then GlimmerM might not predict any thing because it is designed to identify only complete gene models from start to stop To address this issue the authors have developed a new version of GlimmerM called Glimmer MExon which has the ability to predict partial genes Genes that are missing either their 5 or 3 end or both can be recognized by this program GlimmerMExon has thus far only been trained for P falciparum and its perform ance there is comparable to that of GlimmerM on DNA sequences containing at least one com plete gene Currently this prototype lacks an automatic training procedure but the authors plan to adapt GlimmerM s training procedure to work with GlimmerMExon in the near fu ture Acknowledgements This work was supported in part by the National Institutes of Health under grant RO1 LM06845 and by the National Science Foun dation under grants KDI 9980088 and IIS 9902923 Literature Cited Altschul S Gish W Miller W Myers E and Lipman D 1990 Basic local alignment search tool J Mol Biol 215 403 410 Altschul S Madden T Schaffer A Zhang J Zhang Z Miller W a
215. ild 30 which can be found at the Current Protocols Web site http www3 interscience wiley com c_p cpbi_sampledatafiles htm 1 Type the following command at the Unix Linux command prompt firstef lt input file name gt lt output file name gt For example if sequences to be analyzed are stored in a file called example seq type the following command to run FirstEF firstef example seq example result Here the result file is named as example result The user can choose the name of both the input and output files 2 Open the output file using a any text editor APPENDIX IC 3 Analyze results as described see Basic Protocol steps 5 to 9 The main difference between the Web based and local version of FirstEF is the size of the input file The maximum acceptable size of the input file for the Web based version is 100 kb whereas there is no size limit for the local version In fact local FirstEF can analyze any human chromosome sequence all at once However in this version the user has no option to select different cut off values Instead the local version of FirstEF outputs all predictions with P donor 0 3 P promoter 0 3 and P exon 0 5 Please check specific downloaded versions for these cut off values The user must then manually parse these predictions for the cut off values of their choice COMBINING FirstEF PREDICTIONS WITH OTHER ANNOTATIONS As there is a wealth of data in sequence databases GenBank EMBL DDB
216. in Current Protocols in Bioinformatics Finding Genes 4 9 9 Supplement 4 GrailEXP and Genome Analysis Pipeline for Genome Annotation 4 9 10 Supplement 4 could be modified to cluster these EST cDNAs together using GrailEXP s consistency check function Since the primarily goal of Gawain is gene modeling and not EST cDNA clustering its current emphasis is on clustering together the EST cDNAs that support a particular gene model and ignoring the rest COMMENTARY Background Information Gene finding programs Perhaps the first and most fundamental question one might ask of a DNA sequence is whether it is likely to encode the sequence of a protein Several patterns exist in protein coding DNA sequences that are not found in noncoding regions These allow methods to be developed for identifying such regions Early attempts to distinguish coding from noncoding DNA relied upon base composition the triplet nature of genetic code and for some organisms the fre quency of codon usage Fickett 1982 Staden 1984 Fickett and Tung 1992 Such techniques work reasonably well on long gt 200 base cod ing regions such as those found in prokaryotes They are however inadequate for predicting the protein coding regions found in higher eu karyotes which are relatively short 130 base sequences separated by introns that may be hundreds or thousands of bases long A number of protocols have been devised to
217. inclusion of these in the training set GlimmerM will miss these introns because it is designed to find only the standard GT AG introns Setting thresholds for splice site detection Choosing appropriate thresholds for the rou tines that determine splice sites involves decid ing upon a trade off between false negative and false positive rates This can be quite a chal lenging problem when only a small number of true sites are known which is the usual situ ation Invariably there are huge numbers of false positive sites Essentially any GT dinu cleotide that is not a known donor site can be used as a false site and likewise for AG dinu cleotides and donor sites In choosing these thresholds the user must decide how to maxi mize the specificity of the recognition task i e the percentage of false sites that are correctly reject without a big loss in sensitivity i e the number of true sites that are found One auto mated strategy is to set a threshold such that the system will always miss a fixed percentage of the true sites another possibility is to set it so that none of the true sites are missed The latter strategy usually results in a very high false positive rate which causes a serious degrada tion in the overall performance of the gene finder Using a fixed percentage is more appeal ing but the optimal percentage varies from one organism to another Pertea et al 2001 Some times the patterns around a true splic
218. inding in plants Plant Molecular Biology 48 9 48 Pertea M Lin X and Salzberg S L 2001 GeneSplicer A new computational method for splice site prediction Nucleic Acids Res 29 1185 1190 Salzberg S L 1997 A method for identifying splice sites and translational start sites in eukaryotic mRNA Comput Appl Biosci 13 365 376 Salzberg S L Delcher A L Kasif S and White O 1998a Microbial gene identification using interpolated Markov models Nucleic Acids Res 26 544 548 Salzberg S Delcher A L Fasman K H and Hen derson J 1998b A decision tree system for finding genes in DNA J Comput Biol 5 667 680 Salzberg S L Pertea M Delcher A L Gardner M J and Tettelin H 1999 Interpolated Markov models for eukaryotic gene finding Genomics 59 24 31 Stephens R Kalman S Lammel C Fan J Ma rathe R Aravind L Mitchell W Olinger L Tatusov R Zhao Q Koonin E V and Davis R W 1998 Genome sequence of an obligate intracellular pathogen of humans Chlamydia trachomatis Science 282 754 759 Wu Q and Krainer A R 1996 Ul mediated exon definition interactions between AT AC and GT AG introns Science 274 1005 1008 Yuan Q Quackenbush J Sultana R Pertea M Salzberg S L and Buell C R 2001 Rice bioin formatics Analysis of rice sequence data and leveraging the data to other plant species Plant Physiol 125 1166 1174 Key References Salzbe
219. informatics rate than GeneFinder by 13 on the gene level due mainly to better detection of gene boundaries and greater specificity at the exon and nucleotide levels Wei and Brent 2006 N SCAN s results are comparable to TWIN SCAN s R H Brown and M R Brent unpub observ Drosophila species N SCAN is capable of using multiple informant genomes for a single target In the case of Drosophila melanogaster using three other insect species D yakuba D pseudoobscura and A gambia improved the accuracy of gene prediction to 55 gene sensitivity 77 exon sensitivity and 93 nucleotide sensitivity R H Gross and M R Brent unpub observ A later run included D ananassae rather than A gambia and im proved exact gene sensitivity by 5 to 60 R H Brown and M R Brent unpub observ Arabidopsis thaliana The state of the art in Arabidopsis gene finding is even better than for worms although the difference is not large TWINSCAN appears to be much more accu rate than GlimmerM GeneMark hmm GEN SCAN and GeneSplicer Allen et al 2004 and references therein N SCAN has not been optimized for Arabidopsis Factors affecting accuracy Although accuracy on any genome or clade can be improved by tailored modeling and fine tuning gene finding seems to be inherently harder in some genomes than others This is not well understood but intron length and av erage number of introns per transcript are sig nificant
220. ing The orientation of the arrows in the geneid track denoting the gene is annotated in the forward strand USING EXTERNAL INFORMATION TO SOLIDIFY geneid PREDICTIONS One of the strengths of geneid is that it can easily incorporate external information about gene features on the input query sequence in the final gene prediction As human genomic sequences are being annotated with increasing reliability this option may be useful e g to analyze in detail apparently void genomic regions lying between known genes to explore the possibility of alternative exons in known genes with well established consti tutive exonic structure or to extend gene predictions based on partial EST sequences This external evidence can include known exons genes or simply regions highly suspected of coding for proteins In such cases geneid will predict a gene structure compatible with the external information provided The external information can also be a set of candidate exons obtained using some other exon prediction approach computational or experimental In this case geneid will assemble the gene prediction by maximizing the sum of the scores of the assembled exons In any case the gene features to be used by geneid as external information must be provided as GFF files The following describes two examples for which external information substantially improves geneid predictions Necessary Resources Hardware Unix Linux workstation with at least 256 Mb
221. ing GlimmerM to Find Genes in Eukaryotic Genomes 4 4 20 Heidelberg J F Eisen J A Nelson W C Clayton R A Gwinn M L Dodson R J Haft D H Hickey E K Peterson J D Umayam L Gill S R Nelson K E Read T D Tettelin H Richardson D Ermolaeva M D Vamathevan J Bass S Qin H Dragoi I Sellers P McDonald L Utterback T Fleishmann R D Nierman W C and White O 2000 DNA se quence of both chromosomes of the cholera pathogen Vibrio cholerae Nature 406 477 483 Jelinek F 1997 Statistical Methods for Speech Recognition MIT Press Cambridge MA Murthy S K Kasif S Salzberg S and Beigel R 1993 OC1 Randomized induction of oblique decision trees Proc 11th Natl Conf on Artifi cial Intelligence 322 327 Murthy S K Kasif S and Salzberg S 1994 A system for induction of oblique decision trees J of Artificial Intelligence Res 2 1 32 Nelson K E Eisen J A and Fraser C M 2001 Genome of Thermotoga maritima MSB8 Meth ods Enzymol 330 169 180 Pavy N Rombauts S Dehais P Mathe C Ramana D V Leroy P and Rouze P 1999 Evaluation of gene prediction software using a genomic data set Application to Arabidopsis thaliana sequences Bioinformatics 15 887 899 Pertea M Salzberg S L and Gardner M J 2000 Finding genes in Plasmodium falciparum Na ture 404 34 35 Pertea M and Salzberg S L 2002 Computational gene f
222. ing a pseudogene A hyphen in the protein translation of a gene model indicates a frame shift between exons One or more of these may indicate a pseudogene Additionally if Gawain reports a long mRNA but a short protein translation this may indicate a pseudogene containing several stop codons however this may also indicate an error in the genomic sequence It is not easy to distinguish between actual pseudogenes and errors in the alignment program due to a missed short exon which might cause a frame shift or a wrongly identified stop codon due to sequencing error which confuses the 5 3 UTR finder Open and closed end based gene modeling Gawain can be run in open or closed end gene modeling mode Open ended gene modeling allows partial genes to be predicted at the ends i e it assigns no penalties for genes running off the edges of the sequence This is the ideal way to run the program for contigs known not to begin or end a region Running with closed ends on the other hand will penalize partial genes at the ends of the sequence and will try to close off all gene models This is ideal if one knows that a clone only contains one gene that one wants to examine and one does not want Gawain to predict partial genes on the ends of the sequence Gawain runs with closed ends by default since typically there is almost always an initial or terminal GRAIL Exon Candidate that can be predicted near the beginning or end of a
223. ioinformatics d DN o ingl anera predator mpertea Repergillus train test tenp faste d anoe nane AFI Sequerce length Predicted genes exons Strond Exon Type Terninal Initial Initial Internal In Internal Internal Internal Figure 4 4 9 Sample output of the current version of GlimmerM created by the Basic Protocol The FASTA file used to generate this output is available on the Current Protocols Web site http www3 interscience wiley com c_p cpbi_sampledatafiles htm COMMENTARY Background Information Foundation and assumptions The basis of GlimmerM is a dynamic pro gramming algorithm var3 1 that considers all combinations of possible exons for inclusion in a gene model and chooses the best of all these combinations The possible exon intron com binations are formed after an initial screening of the possible translational start sites and splice sites found in the genome Both these entities are determined with specially designed mod ules based primarily on Markov chains Markov models have been in use for decades as a method for modeling sequences In par ticular they have been remarkable for their success in modeling speech Jelinek 1997 Current Protocols in Bioinformatics Markov models are a natural way of modeling a sequence of events and they translate very directly to DNA sequence data Although other methods are in use Markov models are among the most succe
224. ion P donor A posteriori probability of donor for a given GT CpG Window Boundaries of the CpG window of length 201 if the exon is CpG related otherwise the output reads Non CpG related Rank Rank of the first exon within a cluster Figure 4 7 2 shows output from Example 1 PTF two predicted first exons are separated by lt 1000 bp they are considered as first exon predictions of the same gene and ranked based on a posterior probabilities Current Protocols in Bioinformatics USING LOCAL FirstEF TO PREDICT PROMOTERS AND FIRST EXONS FirstEF can be obtained through the Office of Technology Transfer Cold Spring Harbor Laboratory by logging on to the FirstEF Web site clicking the Research Licenses Instructions link and following the directions provided FirstEF software is freely available to nonprofit research institutions in executable form for Unix and Linux platforms The software package includes a README file with instructions for installing the software Note that a research license agreement will need to be signed Necessary Resources Hardware Computer workstation with Unix or Linux operating system Software FirstEF software http rulai cshl org tools FirstEF Files DNA sequences to be analyzed in FASTA format APPENDIX 1B Sample sequences can be found at the FirstEF Web site The example Example 1 used in this unit is a DNA sequence of length 100 kb from human chromosome 20 chr20 300001 400000 NCBI bu
225. ion of polyadenylation or promoter signals USING THE N SCAN WEB SERVER When N SCAN is run through the Web server the user selects the clade and species from which the target sequence is derived Once a selection is made an informant database is automatically selected and displayed on the Web page The user provides the target sequence either by cutting and pasting from another window or by uploading a sequence file When the user submits the job by clicking the Predict Genes button a submission ID is generated and the user is taken to a page that lists the status of the submission When processing is complete this page changes to display the results and an email notification is sent to the user This email contains a link to the results page which will be available until the user deletes the project Necessary Resources Hardware Computer with an Internet connection Software Internet Explorer version 6 0 or higher Firefox v 1 5 or higher or Safari v 1 3 or higher Files The N SCAN server requires genomic DNA sequence as input Each input should contain a single sequence consisting of the letters ACGTN in upper lower or mixed case All other characters including whitespace characters such as line breaks tabs and carriage returns are ignored Optionally the input may begin with a FASTA APPENDIX 1B file header i e a single line beginning with gt followed by a sequence name comments and a carriage return The head
226. ions are incompatible see next step Predict the gene structure in sequence example3 from the set of remaining exons geneid P param human3iso param O example3 filtered exons samples example3 fa Option O instructs geneid to read the set of predicted exons externally instead of predicting them and assemble the optimal gene structure from this set The resulting prediction appears in Figure 4 3 9D geneid EST3b which is now compatible with the EST3 sequence and which closely resembles yet another isoform Force the prediction of the first exon of the gene by providing the coordinates of the promoter element geneid P param human3iso param R samples example3 promoter gff samples example3 fa gt example3 exons gff Even though geneid predictions on example3 correspond quite well to different isoforms of the same gene in all cases geneid fails to predict the first coding exon of the gene Failing to predict short first coding exons is a feature of geneid as well as of other gene prediction programs With geneid there are a number of ways in which the user can force the prediction of a complete gene starting by a First exon e g by using a gene model which defines see Background Information how to assemble only one gene This examples uses the fact that the default gene model includes a promoter feature see Background Information to provide geneid the coordinates of a promoter element which has been experimentally
227. ions page is a link called My Submissions When you click on My Submissions you will be taken to a page displaying all your previous jobs If you click on a number the results page for that job is shown Examine results on the N SCAN output Web page 7 Check the run parameters on the Submissions page The N SCAN output page is shown in Figures 4 8 3 and 4 8 4 Click the Submission Details button to see the target and informant sequences the N SCAN version and links to the input sequence and parameter files This information can be useful for ensuring Current Protocols in Bioinformatics N SCAN Twinscan Gene Predictor Copy and paste gene prediction Submission 54 Submission Details Current State Quoued Masking Aligning Predicting Compito Submission Details Submitted Wed 21 Mar 2007 12 51 23 1900 Started Wed 21 Mar 2007 12 52 34 1900 Status Predicting NSCAN is now predicting genes You will get an email when it is done Copyright 2007 Questions Comments help mblab wust edu Figure 4 8 2 The Submission page that appears when a sequence has been submitted to the Web server The current status of the job is explained at the bottom N SCAN Twinscan Gene Predictor Copy and paste gene prediction Submission 60 Summary 6 Genes Predicted 0 im l Gene All Transcripts submission 60 fa 001 1 All Proteins submission 60 fa 002 1 Whole GTF submission 60 fa 003 1 Fi
228. ire installation of additional software In cases where the PostScript file was augmented with additional symbols due to the E mail transfer to extract a valid PostScript file one needs to select every line inclusive between PS Adobe 2 0 and EOF and save it to a file with extension ps The file should not contain any blank lines before the PS Adobe 2 0 line or after the EOF line This file can then be viewed with any PostScript viewer program e g GhostView available at http www cs wisc edu ghost Current Protocols in Bioinformatics OSsf nu mn Direct Sequence A Complementary Sequence m Genehlarkhmm prediction Thu Jul 25 10 37 56 EDT 2002 Order 4 Window 96 Step 12 3 10 73 mS a im 7 Figure 4 6 6 The graphical output from the Eukaryotic GeneMark hmm program for a region of the example sequence The six different panels represent the six possible reading frames three each on the direct and reverse strands Current Protocols in Bioinformatics Finding Genes 4 6 7 Supplement 1 ALTERNATE PROTOCOL 1 Eukaryotic Gene Prediction Using GeneMark hmm 4 6 8 Supplement 1 Figure 4 6 6 shows page 3 of the PostScript file generated by the GeneMark hmm Web server for the sample file and viewed in the GhostView viewer Two correctly predicted internal exons 2564 2621 and 4076 4208 are shown as thick black lines The first appears in the first frame of the dir
229. irst frame is more likely to be the real one because Fr1 is larger than Fr2 Although MZEF does not assemble the exons into a gene model occasionally one can resolve the frame ambiguity or eliminate the false positive exon predictions by requiring frame compatibility between adjacent coding exons In the Web example above see Basic Protocol 1 Figure 4 2 1 the predicted exon 6802 6934 had two ORFs i e Orf 211 with Fr2 0 553 Fr3 0 522 but in order for it to be compatible with the adjacent coding exons the second ORF would have to be used For similar reasons the predicted exon 13341 13425 may be a false positive because its ORF is not compatible with others and its P score is relatively low compared to that of the adjacent ones One must be careful when using frame compatibility because it assumes the adjacent ones are correct and there is no missing false negative one next to it Sometimes a true exon s frame is not compatible to the next predicted one because of alternative splicing i e it may be compatible with another one further downstream Current Protocols in Bioinformatics Finding Genes 4 2 9 Using MZEF to Find Internal Coding Exons 4 2 10 COMMENTARY Background Information MZEF is based on Quadratic Discriminant Analysis QDA QDA assumes real exons and pseudoexons are distributed as two different normal distributions in the feature space it uses training data to construct the optim
230. is useful to consider sensitivity and specificity separately Sensitivity is the number of correctly predicted features as a fraction of the true number in the sequence Specificity is the number of cor rectly predicted features as a fraction of the number predicted For example exact gene sensitivity is the number of genes predicted exactly right start codon stop codon and all splice sites in between divided by the num ber of genes present in the sequence Different measures of accuracy are appropriate for dif ferent applications For example in a project aimed at cloning thousands of full length open reading frames pursuing genes whose bound aries have been miscalled is expensive Thus a gene predictor with high specificity for both the start and stop codons is appropriate At the nucleotide and exon levels there is often a sensitivity specificity tradeoff with the most sensitive predictor doing poorly on specificity and vice versa For example sensitivity at the nucleotide level can be increased at the cost of specificity by simply classifying randomly selected nucleotides as coding Some of these will in fact be coding nucleotides increasing sensitivity while others will not decreasing specificity Predicting a complete gene cor rectly by such a simple procedure however is extremely unlikely For example either over predicting or underpredicting of starts or stops will limit both sensitivity and specificity on the
231. it good exor Grail good exon Grol excelent exom Groit good 4 gt exon MZEF P 0 990 exon MZEF P 0551 exom NZEF P0750 exom NZEF P0333 exom Groit excelent b 4 enon MZEF P O437 enon Grow excefient evos Croit excelent exor Grot good evon Crait marginal 4 gt 4 4 gt evom Groit excelent exom MZEF P0999 enom Grok excelent enom Grail excelent gt 4 gt evom MZEF P0307 evon MZEF P 0 577 evom Grail excelent b b exon Croat excelent gt lt anas ie i tre EA MER _ te pan b b kd gt Retrieve 13 Alignment BLASTN Figure 4 1 3 Annotated output from GeneMachine showing the results of multiple gene prediction program runs NCBI Sequin is used as the viewer At the top of the output are shown the results from various BLAST runs BLASTN vs dbEST BLASTN vs nr and BLASTX vs SWISS PROT Towards the bottom of the window are shown the results from the predictive methods FGENES GENSCAN MZEF and GRAIL 2 Annotations indicating the strength of the prediction are preserved and shown wherever possible within the viewer Putative regions of high interest would be areas where hits from the BLAST runs line up with exon predictions from the gene prediction programs Sequin as a workbench and graphical viewer Using Sequin also has the advantage of presenting the results to the user in a familiar format basically the same format that is used in Entrez for graphical views The most
232. ith a URL of the form http compbio ornl gov GAT_tmp req_id html where req_id represents the Request ID The user may type this URL into the browser s Address Internet Explorer or Location Netscape box at any time to check on the progress of the request 12 Examine the results returned in the browser window USING GENOME ANALYSIS PIPELINE FOR COMPREHENSIVE ANALYSIS OF DNA SEQUENCES Several Genome Analysis Pipelines can be accessed from the ORNL Genome Pipeline Web site http compbio ornl gov genomepipeline i e Eukaryotic Human and Mouse Eukaryotic Yeast and Prokaryotic This discussion focuses on the Eukaryotic Human and Mouse pipeline The first control on this pipeline Web form Fig 4 9 2 is a pull down menu labeled Select organism which allows the user to select the organism of interest The next control is acheck box labeled Select all services This control is provided for the convenience of selecting or deselecting all supported analysis services on the form with a single mouse click It is also possible to select or deselect the individual supported analysis services on the form For each selected service there may be one or more parameter options that can be selected or set The Eukaryotic Human and Mouse Genome Analysis Pipeline Fig 4 9 2 supports several analysis tools The supported gene finders are GrailEXP and GENESCAN The user can specify post processing analysis on the predicted g
233. kb are readily analyzed using the Web RepeatMasker with the time needed for processing correlating with the length of the sequence For faster service outside North America there are RepeatMasker mirror sites in Germany Israel and Australia On the other hand if one routinely submits large sequences for analysis it may be better to download the command line version and run RepeatMasker locally see Basic Protocol 2 Importantly if the query sequence exceeds the 100 kb limit the only choice is to download RepeatMasker and run it locally Necessary Resources Hardware Any Internet connected computer Software Web browser e g Mozilla Firefox or Internet Explorer Files A FASTA file APPENDIX 1B or a collection of FASTA files can be processed via the Web interface Note that the size limit is 100 kb for RepeatMasker via Web The example file used in this protocol is a 22 539 bp human genomic DNA sequence from the UCSC Genome Browser http genome ucsc edu cgi bin hgGateway The coordinate is chr10 62743355 62765893 1 Point the Web browser to http Avww repeatmasker org cgi bin WEBRepeatMasker Load the FASTA sequence file maximum 100 kb by entering the sequence name or browsing the file Alternatively paste the FASTA sequence maximum 100 kb into the indicated text field RepeatMasker will return an error message if the input sequence contains non DNA symbols or if the sequence is too long 2 Select a format for re
234. lation start codon ATG and the final exon ends at the position containing the stop codon for that gene Previous studies showed that P falciparum exons have an average AT content of 70 to 75 Salzberg et al 1999 so GlimmerM also displays the A T percentage for each exon If a model overlaps another gene model that has a higher coding score or is simply longer but either the score difference or the overlap length is small then the message Bad overlap with gene x is printed where x is the ID number of the better overlapping model Other Organisms For all versions of GlimmerM except the malaria version the system prints a list of the putative gene models Fig 4 4 9 The output is very similar to that produced by the GENSCAN system Burge and Karlin 1997 this format is intentionally designed to make it easier for software that parses the output of GENSCAN to use GlimmerM as well For each gene model the output contains a list of the exons that comprise that prediction Four types of exons may appear in the predictions initial between a start codon and a donor site internal between two splice sites terminal between an acceptor site and a stop codon and single for unspliced genes The exon length is also printed for reference No probabilities are assigned to the exons but this feature in a subsequent version of GlimmerM is planned to be implemented in order to permit the user to identify the best exons Current Protocols in B
235. li V Katinka M Vacherie B Biemont C Skalli Z Cattolico L Poulain J De Berardinis V Cruaud C Duprat S Brottier P Coutanceau J P Gouzy J Parra G Lardier G Chap ple C McKernan K J McEwan P Bosak S Kellis M Volff J N Guig R Zody M C Mesirov J Lindblad Toh K Birren B Nusbaum C Kahn D Robinson Rechavi M Laudet V Schachter V Quetier F Saurin W Scarpelli C Wincker P Lander E S Weissenbach J and Roest Crollius H 2004 Genome duplication in the teleost fish Tetraodon nigroviridis reveals the early vertebrate proto karyotype Nature 431 916 917 Lewis S E Searle S M J Harris N Gibson M Iyer V Ricter J Wiel C Bayraktaroglu L Birney E Crosby M A Kaminker J S Matthews B Prochnik S E Smith C D Tupy J L Rubin G M Misra S Mungall C J and Clamp M E 2002 Apollo A se quence annotation editor Genome Biology 3 re search0082 Mott R 1997 EST GENOME A program to align spliced DNA sequences to unspliced genomic DNA Comp Appl Biosci 13 477 478 Mouse Genome Sequencing Consortium 2002 Ini tial sequencing and comparative analysis of the mouse genome Nature 420 520 562 Finding Genes 4 3 27 Supplement 18 Using geneid to Identify Genes 4 3 28 Supplement 18 Parra G Blanco E and Guig6 R 2000 geneid in Drosophila Genome Res 10 511 515 Parra
236. limmerM should be used as one of a suite of tools Accurate gene identification depends on using every tool available and the description in this unit should not be taken as an implication that GlimmerM alone can find all genes in a given genome In order to produce reasonably accurate gene annotations any comprehensive annotation effort needs several computational tools e g searches using BLAST units 3 3 amp 3 4 and or PSI BLAST against a non redundant protein sequence database Altschul et al 1990 1997 gapped alignments of DNA to protein and EST sequence databases Florea et al 1998 prediction of putative signal peptides tools to detect frame shift errors e g Framesearch UNIT 3 2 and graphical tools to allow annotators to view all the evidence concurrently When no database matches or other computational evidence are found to support a GlimmerM prediction then just as for any other gene finder further investigation is required to confirm these models At the request of the annotation team the GlimmerM system trained for malaria was designed to produce multiple gene models for some genes In all other current versions of GlimmerM only the highest scoring model of a putative gene appears in the output The discussion below explains how to interpret the output of GlimmerM in the special case when overlapping gene models are predicted Although many of GlimmerM s predictions are likely to be correct it is undoubtedly the case
237. ll of the organism specific versions found in version 1 2 however the performance of these versions is slightly different due to changes in parameter settings when building the later system GlimmerM is available free of charge to researchers using it for non commercial purposes The system includes source code and a readme file describing how to compile and train the system In order to obtain the system a representative of a nonprofit organization should fill out a license agreement available on the TIGR Web site http www tigrorg under Software Interested commercial organizations should see the Web site for additional instructions For nonprofit organizations the system is made available almost immediately after submitting the license agreement Files Format the training data in two files a A single FASTA file APPENDIX 1B containing all the DNA sequences for the training data e g gt Seql DNA sequence containing one or more genes AGTCGTCGCTAGCTAGCTAGCATCGAGTCTTTTCGATCGAGGACTAGA CTAGCTAGCTAGCATAGCATACGAGCATATCGGTCATGAGACTGATTGGGGTGTGTGC TAAACTGTGT gt Seq2 another DNA sequence containing more genes TTTAGCTAGCTAGCATAGCATACGAGCATAT CGGTAGACTGATTGGGTTTATGCGTTA b A file specifying the locations of the known genes by the coordinates of the coding portions of those genes in each sequence in the FASTA file For each coding exon its 5 and 3 ends should be listed in order from start to stop Thus genes on the complementary
238. lts for details Adjust speed by selecting among the four radio buttons next to Speed Sensitivity rush quick default or slow Note that a faster speed is associated with a lower sensitivity For example shown here se lect default for Speed Sensitivity See Guidelines for Understanding Results for details Current Protocols in Bioinformatics Summary file name RM2sequpload_1212744700 sequences 2 total length 22539 bp 22539 bp excl N X runs GC level 35 84 amp bases masked 10789 bp 47 87 number of length percentage elements eccupied of sequence SINEs 14 2241 bp 9 94 ALUs 0 0 bp 0 00 8 MIRs 14 2241 bp 9 94 LINEs 10 7375 bp 32 72 amp LINE1 9 7071 bp 31 37 amp LINE2 1 304 bp 1 35 L3 CR1 e 0 bp 0 00 amp LTR elements 0 0 bp 0 00 amp MaLRs 0 0 bp 0 00 ERVL 0 0 bp 0 00 amp ERV _classI ts 0 bp 0 00 amp ERV_classII 0 0 bp 0 00 DNA elements S 1079 bp 4 79 MER1_type 3 585 bp 2 60 amp MER2_type 1 222 bp 0 98 Unclassified 0 0 bp 0 00 Total interspersed repeats 10695 bp 47 45 Small RNA 0 0 bp 0 00 amp Satellites a 0 bp 0 00 amp Simple repeats 1 42 bp 0 19 amp bp 0 23 Low complexity 1 52 most repeats fragmented by insertions or deletions have been counted as one element Figure 4 10 4 Web RepeatMasker result from an example run showing the Summary section which summarizes and categorizes repetitive elements fo
239. luded to above when discussing the performance statistics the reader must keep in mind that these statistical measures were computed at a given point in time The scientists who have developed these algorithms continue to refine and improve their approaches as more and more biological evidence becomes available improving the predictive power of their methods In addition altogether new methods are introduced as well methods that obviously would not be part of any previously published comparisons The most recent example of a new method being introduced is FirstEF developed by Michael Zhang s group at the Cold Spring Harbor Laboratory Davuluri et al 2002 UNIT 4 7 This method uses knowledge of experimentally validated first exons and promoters from gt 2000 genes in order to accurately predict promoter regions first exons The develop ment of FirstEF unir 4 7 provides an excellent example of how the near completion of the human genome has helped to increase our understanding of gene structure Further more new twists on old methods have greatly increased the power of these methods an Current Protocols in Bioinformatics Finding Genes 4 1 5 Supplement 6 An Overview of Gene Identification 4 1 6 Supplement 6 example of this is the GeneMark suite of programs UNITS 4 5 amp 4 6 The latest version of GeneMark hmm uses Hidden Markov Models of protein coding and noncoding se quences and information on start and
240. m acceptor site score 0 38 Minimum donor site score 0 26 Minimum total splice site score acceptor site score donor site score 0 79 The purpose of setting such thresholds is to reduce the amount of false positives and to cut down CPU time perhaps at a reasonable expense of a few false negatives Finally MZEF can only output exons that have a P value gt 0 5 Most often the troubleshooting should start by checking if the input sequence file format is correct FASTA format APPENDIX 1B One should always check the sequence length in the output report and see if it is correct If it is not correct it is most likely caused by extra blank spaces or gt 80 character per line in the sequence file One should always test the program with a gene of known structure If the number of predicted exons is too small try to increase the PO and vice versa Suggestions for Further Analysis One should always run several gene finding programs such as GENSCAN FGENES GRAIL and others Extensive research has shown that an exon predicted with a high score from more than two programs is most likely real even if there is no cDNA support Current Protocols in Bioinformatics because the exon may only be expressed under special conditions Homology searches against known gene databases are also indispensable MZEF should also be run in conjunction with other programs that can predict different types of exons and or different parts of the gene
241. me using an alternative alignment program such as WU BLAST Further the advantages limitations and known bugs of the software are discussed Finally guidelines for understanding the results are provided Curr Protoc Bioinform 25 4 10 1 4 10 14 2009 by John Wiley amp Sons Inc Keywords RepeatMasker e genome annotation e repetitive elements e repeat library e cross_match e WU BLAST e RECON INTRODUCTION RepeatMasker developed by A RA Smit R Hubley and P Green see http www repeatmasker org was designed to identify and annotate repetitive elements in nucleotide sequences and mask them for further analysis The repetitive elements including low complexity DNA sequences and interspersed repeats are annotated and replaced by Ns Xs or lowercase letters see below for options in the correspond ing positions of the DNA sequence The new addition to the RepeatMasker package is a program that also identifies repetitive elements within protein sequences Here we focus on utilizing RepeatMasker to identify repetitive elements in genomic se quences To run RepeatMasker one needs to select the repeat library files which contain repetitive elements consensus sequences Currently Repbase Update Jurka 2001 Jurka et al 2005 http Avww girinst org is the largest commercially avail able repeat library free for academic use and covers a number of organisms in cluding human rodent zebrafish Drosophila and Arabidopsis t
242. mended and Internet access Software An up to date Internet browser such as Internet Explorer http www microsoft comlie Netscape http browser netscape com Firefox http www mozilla org firefox or Safari Attp www apple com safari Current Protocols in Bioinformatics Files None To obtain the geneid software from the Web go to the Distribution section of the geneid Web page and click on Full Distribution Current release geneid v1 2 To obtain the geneid software by anonymous FTP run an FTP session as follows sftp genome imim es Name anonymous Password ftp gt cd pub software geneid ftp gt binary ftp gt get README ftp gt get geneidvl1 2 March_1_2005 tar gz ftp gt quit The geneid distribution has been compressed in a single file geneid_v1 2 March_1_2005 tar gz using the Linux command tar zcvf To uncompress and extract the files type the following commands sgzip d geneid v1 2 March 1 2005 tar gz star xvf geneid v1 2 March1 2005 tar On Linux systems type Star zxvf geneidv1l 2 March_1_2005 tar gz After uncompressing the geneid distribution the directory geneid will have been created in the current working directory The geneid directory contains several subdi rectories and files docs geneid documentation include geneid header file param Parameter files for several organisms samples FASTA sequences used in this unit src geneid source code GNU License M
243. mer and Borodovsky 1999 Besemer et al 2001 COMMENTARY Background Information GeneMark The GeneMark gene prediction algorithm was developed in several steps The first step was performed by a group at the Institute of Molecular Genetics in Moscow in 1986 In a series of three publications it was demon strated that inhomogeneous Markov chain models were useful tools for DNA sequence analysis and particularly for gene prediction Borodovsky et al 1986a b c The GeneMark method itself was described in 1993 Boro dovsky and McIninch 1993 Finding unno ticed genes in the E coli DNA sequences by using GeneMark Borodovsky et al 1994a b served as solid evidence for the accurate pre dictive ability of the method Another advan tage of the method is its flexibility in a sense of rather easy training for a new species or even for a separate class of genes within a given genome Borodovsky et al 1995 Application of the GeneMark program to help in interpret ing the genomic sequences of H influenzae and M genitalium have taken advantage of these features of the algorithm Fleischmann et al 1995 Fraser et al 1995 Tatusov et al 1996 The next step was necessary when com pletely new genomes M jannaschii and H pylori entered the scene with no experimen tally studied segments that could be used for training Bult et al 1996 Tomb et al 1997 At this point a new routine called GeneMark Genesis was de
244. mm is gmhmme The program requires at least two parameters the name of the DNA sequence file and the name of a matrix file supplied after the m option The latter contains parameters of statistical models for DNA sequence analysis generated from a training set of reliably annotated sequences from a particular organism The species for which the matrix file was built must match the name of the species of the DNA sequence origin By default the output is saved in a file named after DNA sequence file with addition of the 1st extension Option o allows users to specify the output file name different from the default name An example would be sgmhmme m12523 fna m human mtx o m12523 1st 1b Alternatively if GeneMark hmm predictions are run routinely for the same organism the matrix file name and the path to the file can be specified in the environmental variables DEFMAT_HMME and MATPATH respectively In the Unix csh and ksh shells this can be done as shown in Figure 4 6 7 The matrix specifications can then be omitted on the command line If using the sample sequence provided the command will be as follows sgmhmme m12523 fna o m12523 1st The file m12523 1st will contain the output of the GeneMark hmm program Current Protocols in Bioinformatics ssetenv DEFMAT HMME human gt export DEFMAT HMME human mtx mtx setenv MATPATH home GeneMark hmm matrices gt export MATPATH home GeneMark hmm matrices csh csh
245. n There are two ways in which a user can analyze sequence data using MZEF One option is to access the MZEF Web interface see Basic Protocol 1 The other is to download and install the Unix version of MZEF which can be run interactively see Basic Protocol 2 or from the command line see Alternate Protocol USING MZEF TO ANALYZE GENOMIC DNA SEQUENCES VIA THE WEB INTERFACE MZEF may be accessed through the Web at http www cshl edu genefinder A user can select the Human Mouse or Arabidopsis the Fission Yeast button would lead to a different algorithm POMBE see Chen and Zhang 1998 buttons and obtain a brief description README file by clicking the link at the bottom of the page Different organism options are available since the rules for gene finding vary slightly from organism to organism In the case of fission yeast the user is redirected to POMBE a linear discriminant analysis based method developed by T Chen The program provides exon predictions on yeast data Since there is no MZEF version for yeast the link to POMBE is provided for the user s convenience Once the selection is made a request form will be generated through which the prediction can be submitted Necessary Resources Hardware For Web access any internet connected computer Software A Web browser Files A FASTA file APPENDIX 1B with no more than 80 characters per line that contains the DNA sequence maximum 200 kb in which one wishes
246. n chromosome 21 coordinates 13 903 812 13 935 812 This sequence can also be found at the samples subdirectory within the geneid distribution see Support Protocol and on the Current Protocols in Bioinformatics Web site at http www currentprotocols com NOTE In a Unix system the syntax to use geneid is geneid options P parameter_file input_sequence where parameter_file isa file containing gene model parameters for a given species or taxonomic group which the user normally downloads with the geneid distribution and input_sequence is a file containing a DNA sequence in FASTA format APPENDIX 1B A number of options allow modification of the geneid default behavior The following assumes that geneid has been successfully installed in a directory within the file system the geneid directory and that this directory is the current working directory see Support Protocol NOTE An introduction to the Unix environment can be found in APPENDIX IC 1 Run geneid on the first example examplel fa with default options geneid P param human3iso param samples examplel fa geneid is a Unix command line program that requires as input a file containing a DNA se quence in FASTA format samples examplel fa see APPENDIX 1B for discussion of FASTA format and a parameter file This is specified by using the option P followed by the name of the parameter file geneid provides parameter files for human this example Drosophila melanog
247. n such a computer Alternate Protocol 2 describes the use of a Perl script which can automate the operation of N SCAN on a local computer Current Protocols in Bioinformatics 4 8 1 4 8 16 December 2007 Published online December 2007 in Wiley Interscience www interscience wiley com DOI 10 1002 0471250953 bi0408s20 Copyright 2007 John Wiley amp Sons Inc UNIT 4 8 Finding Genes 4 8 1 Supplement 20 BASIC PROTOCOL Using N SCAN or TWINSCAN 4 8 2 Supplement 20 N SCAN requires genomic sequences from at least two organisms of appropriate evo lutionary distance Currently N SCAN is available for sequences from the following clades Mammalia mammals Caenorhabditis a genus of roundworms including C elegans and C briggsae Drosophila fruit files and Cryptococcus neoformans a species of pathogenic fungus including strains JEC21 and H99 An earlier version of N SCAN TWINSCAN is available for Brassicaceae mustard family including Ara bidopsis thaliana Zea maize and Oryza rice For all these clades the N SCAN Web server provides the appropriate informant database N SCAN is tuned to predict the 5 untranslated region and the coding portions of genes as accurately as possible with as few false positives as possible It is designed to annotate large sequences such as whole genomes automatically N SCAN cannot predict multiple splices for a single gene and is not tuned for accurate predict
248. n4 E h x l Vu v and 0 7 Var h x V Y V the most popular choice for the optimal V is the following Equation 4 2 5 1 1a y v 32 32 H H 4 2 5 Using MZEF to Find Internal Coding Exons 4 2 16 Current Protocols in Bioinformatics which maximizes the Fisher criterion n _ 6 0_ Fisher 1936 One notices that the Fisher coefficient Equation 4 2 5 will reduce to that of Equation 4 2 4 when x _ although minimization of the Fisher criterion cannot provide an optimal value for the constant threshold v which may be chosen by minimizing the classification errors in the linear subspace Using a linear discriminant function often the Fisher discriminant function for classification is called LDA linear discriminant analysis see Solovyev et al 1994 In real applications one normally does not know the distributions One should always try to transform variables so that they are approximately normal there are many techniques for doing this for instance the Box Cox transformation 1964 Even if one assumes some parametric distributions estimation of the parameters using the training data is still necessary LDA is more robust because it does not require normality of the distributions and it has fewer parameters to be estimated But if one has sufficient data and the decision boundary is intrinsically nonlinear two class distributions have very different shapes as indicated by x X_ QDA may
249. nalysis all nonalphabet symbols are ignored and all ambiguous letters other than the Current Protocols in Bioinformatics D Genebiaric Heuristic Approach Microsoft internet Explorer Ge CR ye frae be bb Q O 2 d jH Favertes QB rete 7 E hte iopal taiog pach edulGerettet harutee hem op Googie euhm Geah rh Heuristic Approach for Gene Prediction in wet ph danse Raloed thes pose Reference Besemer and Borodovsky M Hex 1999 Vol 27 No 19 pp 3911 3920 Download PDF The models used by GeneMark hmm 2 0 and GeneMark 2 4 are derived from parameters measured from the input sequences and knowledge gained through the study of various bacterial genomes These models have been shown to accurately predict in bacterial viral and plasmid ONA sequences Please note that email is the only way to receive output for sequences larger than 1 MB Web site has been redesigned and moved 4 to new more powerful server Input Sequence Title Coptional e Sequence o Sequence File upload e Use alternate genetic code o O Mycoplasma TGA Trp Output Options Email Address required for graphical output or sequences longer than 1000000 bp e O Generate PostScript graphics Print GeneMark 2 4 predictions in addition to GeneMark hmm predictions O Translate predicted genes into protein Detour Figure 4 5 7 The user interface for the Heuristic model building program Required input includes a DNA seque
250. nce either copied and pasted into the text box or uploaded as a file from the users computer in FASTA program builds a model from the sequence and then runs GeneMark hmm symbols of the four nucleotides assuming that they occur rarely are replaced with C This minimizes the chance of the possible creation of a false start or stop codon 2 Scroll down the page to the Use Alternative Genetic Code check box and if the species to be analyzed uses an alternative genetic code Mycoplasma species select this option 3 Scroll further down the page and set the Output Options The user may request the graphical output Also GeneMark predictions can be requested in addition to ones of GeneMark hmm An E mail address is required for sending text output for sequences longer than 1 Mbp or if graphical output is requested 4 After completing the above entries click the Start GeneMark hmm button The results will be depicted on the browser or will be sent to the E mail address provided Current Protocols in Bioinformatics format This Finding Genes 4 5 11 Supplement 1 ALTERNATE PROTOCOL 2 Prokaryotic Gene Prediction Using GeneMark and GeneMark hmm 4 5 12 Supplement 1 5 Interpret the text output The text output from the Heuristic approach is identical to that of GeneMark hmm see Alternate Protocol 1 6 Interpret the graphical output The graphical output from the Heuristic approach is identical to
251. nd Lipman D 1997 Gapped blast and psi blast A new generation of protein database search programs Nucleic Acids Res 25 3389 3402 Current Protocols in Bioinformatics The Arabidopsis Genome Initiative 2000 Analysis of the genome sequence of the flowering plant Arabidopsis thaliana Nature 408 796 815 Bowman S Lawson D Basham D Brown D Chillingworth T Churcher C M Craig A Davies R M Devlin K Feltwell T Gentles S Gwilliam R Hamlin N Harris D Hol royd S Hornsby T Horrocks P Jagels K Jassal B Kyes S McLean J Moule S Mun gall K Murphy L Barrell B G et al 1999 The complete nucleotide sequence of chromo some 3 of Plasmodium falciparum Nature 400 532 538 Brendel V and Kleffe J 1998 Prediction of locally optimal splice sites in plant pre mRNA with applications to gene identification in Arabidop sis thaliana genomic DNA Nucleic Acids Res 26 4748 4757 Burge C 1997 Ph D thesis Identification of Genes in Human Genomic DNA Standford University Calif Burge C B and Karlin S 1997 Prediction of com plete gene structures in human genomic DNA J Mol Biol 268 78 94 Dietrich R C Incorvaia R and Padgett R A 1997 Terminal intron dinucleotide sequences do not distinguish between U2 and U12 dependent introns Mol Cell 1 151 160 Florea L Hartzell G Zhang Z Rubin G M and Miller W 1998 A computer program for ali
252. nd clinical research in the future ACKNOWLEDGMENTS This unit was written by Dr Andreas D Baxevanis in his private capacity No official support or endorsement by the National Institutes of Health or the United States Depart ment of Health and Human Services is intended or should be inferred LITERATURE CITED Altschul S F Madden T L Schaffer A A Zhang J Zhang Z Miller W and Lipman D J 1997 Gapped BLAST and PSI BLAST A new generation of protein database search programs Nucl Acids Res 25 3389 3402 Burset M and Guig R 1996 Evaluation of gene structure prediction programs Genomics 34 353 367 Chothia C and Lesk A M 1986 The relation between the divergence of sequence and structure in proteins E M B O J 5 823 826 Claverie J M 1997a Computational methods for the identification of genes in vertebrate genomic sequences Hum Mol Genet 6 1735 2744 Claverie J M 1997b Exon detection by similarity searches Methods Mol Biol 68 283 313 Claverie J M 1998 Computational methods for exon detection Mol Biotechnol 10 27 48 Davuluri R V Grosse I and Zhang M Q 2002 Computational identification of promoters and first exons in the human genome Nature Genetics 29 412 417 Guig6 R 1997 Computational gene identification J Mol Med 75 389 393 Guig6 R Knudsen S Drake N and Smith T 1992 Prediction of gene structure J Mol Biol 226 141 257 Harris N L 1997 Genotator
253. ne recognition for both DNA strands Comput Chem 17 123 133 Borodovsky M Sprizhitsky Yu Golovanov E and Alexandrov A 1986a Statistical patterns in primary structures of functional regions in the E Current Protocols in Bioinformatics coli genome I Oligonucleotide frequencies analysis Mol Biol 20 826 833 Borodovsky M Sprizhitsky Yu Golovanov E and Alexandrov A 1986b Statistical patterns in primary structures of functional regions in the E coli genome II Non homogeneous Markov models Mol Biol 20 833 840 Borodovsky M Sprizhitsky Yu Golovanov E and Alexandrov A 1986c Statistical patterns in primary structures of functional regions in the E coli genome III Computer recognition of cod ing regions Mol Biol 20 1145 1150 Borodovsky M Rudd K and Koonin Eu 1994a Intrinsic and extrinsic approaches for detecting genes in a bacterial genome Nucleic Acids Res 22 4756 4767 Borodovsky M Koonin Eu and Rudd K 1994b New genes in old sequences A strategy for find ing genes ina bacterial genome Trends Biochem Sci 19 309 313 Borodovsky M McIninch J Koonin E Rudd K Medigue C and Danchin A 1995 Detec tion of new genes in the bacterial genome using Markov models for three gene classes Nucleic Acids Res 23 3554 3562 Bult C J White O Olsen G J Zhou L Fleischmann R D Sutton G G Blake J A FitzGerald L M Clayton R A Gocayne
254. ng and neural networks Nucl Acids Res 21 607 613 Snyder E E and Stormo G D 1997 Identifying genes in genomic DNA sequences Jn DNA and Protein Sequence Analysis M J Bishop and C J Rawlings eds pp 209 224 Oxford University Press New York Stormo G D 2000 Gene finding approaches for eukaryotes Genome Res 10 511 515 Wevrick R Kerns J A and Francke U 1996 The IPW gene is imprinted and is not expressed in the Prader Willi syndrome Acta Genet Med Gemollol 45 191 297 Zhang J and Madden T L 1997 PowerBLAST A new network BLAST application for interactive or automated sequence analysis and annotation Genome Res 7 649 656 Contributed by Andreas D Baxevanis National Human Genome Research Institute National Institutes of Health Bethesda Maryland Current Protocols in Bioinformatics Finding Genes 4 1 9 Supplement 6 Using MZEF to Find Internal Coding Exons MZEF Michael Zhang s Exon Finder Zhang 1997 was designed to help identify one of the most important classes of exons i e internal coding exons in human genomic DNA sequences Zhang 1998c It is neither for predicting intronless genes nor for assembling predicted exons into complete gene models There is also a mouse version mMZEF and an Arabidopsis version aMZEF and they can all be found at http www cshl edu genefinder Since they all have the same interface this unit will only describe how to use the human versio
255. nly it does not incorporate any Current Protocols in Bioinformatics database or homology information Users wishing to emulate the behavior of the old GRAIL 1 3 system should select this option Also see Background Information for discus sion of Perceval 5 Choose whether or not to perform a search of gene message databases and which databases to search Checking the box labeled Galahad EST mRNA cDNA Alignments will run Galahad GrailEXP s BLAST based alignment algorithm using the selected databases By default the GrailEXP database which incorporates a number of publicly available databases is selected but the user can narrow the search by deselecting this database and selecting one or more component databases The output from this program will be putative exon boundaries determined from the alignment with the ESTs mRNAs and or cDNAs in the target databases Also see Background Information for discussion of Galahad 6 Choose whether or not to assemble the requests from steps 4 and 5 into complete gene structures Checking the box labeled Gawain Gene Predictions will run Gawain GrailEXP s dynamic programming gene assembly algorithm which will assemble complete gene structures from the neural net predicted exon candidates if requested in step 4 and the alignment based exon candidates if requested in step 5 By default Gawain uses all the alignments obtained from Galahad However the Gene modeling organism options
256. no tation is as follows CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 18526 18555 CDS join 1776 1854 2564 2621 4076 4208 6041 6252 6802 6934 7759 7856 9444 9573 10867 11081 12481 12613 13702 13799 14977 15115 15534 15757 16941 17073 17688 17732 that may be compared with the MZEF predictions below The FORTRAN program also requires the following data files which are available from the FTP site see steps 1 and 2 below asl as2 Pri br2 dsl ds2 dat dat dat dat dat dat h6 ex1 dat h6 ex2 dat h6 excl h6 exc2 h6 exil h6exi2 h exll h6ex12 h exrl h6 exr2 qda dat dat dat dat dat dat dat dat dat and test dat is just a short input DNA sequence for a test run NOTE The names of the data files for each organism are the same but the contents of the files differ 1 Create a new directory to hold the MZEF files and change to that directory mkdir MZEF cd MZEF Current Protocols in Bioinformatics Coordinates 1817 1854 4076 4208 6041 6252 6072 6252 6802 6934 7759 7856 9444 9573 9449 9573 10867 11081 10914 11081 12481 12613 12505 12613 13341 13425 13357 13425 13702 13799 13730 13799 14977 15115 15534 15757 16941 1707
257. nsider them as probable novel exons If the cDNA EST alignments have different splice patterns keep them separate 5 Merge each of the assembled transcripts with a FirstEF predicted cluster that falls on the same strand If a predicted first exon overlaps with the first exon of assembled transcript from step 4 then consider that as the first exon of the corresponding transcript Otherwise assign the nearest 5 first exon prediction to the assembled transcript Figure 4 7 3 shows the assembled transcripts and first exon predictions for the DNA sequence that spans 1 to 60000 bp in Example 1 Table 4 7 2 presents the annotations and their corresponding exon coordinates Current Protocols in Bioinformatics Finding Genes 4 7 7 Supplement 1 Application of FirstEF to Find Promoters and First Exons in the Human Genome 4 7 8 Supplement 1 GUIDELINES FOR UNDERSTANDING RESULTS Like any gene prediction program based on pattern recognition methods FirstEF has false positive i e a genomic region that is not a real exon is predicted as real exon and false negative i e a genomic region that is in fact a real exon is not predicted by FirstEF predictions The user should experiment with different cut off probability values and consider three important points in interpreting the results of FirstEF CpG Related and Non CpG Related First Exons The accuracy of FirstEF is higher for CpG related first exons with an a
258. nsitions between hid den states The HMM framework of GeneMark hmm i e the logic of transitions between hidden Markov states followed the logic of the genetic structure of the bacterial genome The Markov models of coding and noncoding regions were incorporated into the HMM framework to gen erate stretches of DNA sequence with coding or noncoding statistical patterns This type of HMM architecture is known as HMM with duration The sequence of hidden states asso ciated with a given DNA sequence carries in formation on positions where coding function is switching into noncoding and vice versa The sequence of hidden states constitutes the HMM trajectory The core GeneMark hmm proce dure a dynamic programming type algorithm finds the most likely HMM trajectory given the DNA sequence The newest version of Gene Current Protocols in Bioinformatics Mark hmm Besemer et al 2001 has the ca pability of predicting genes with overlaps of arbitrary length This version also integrates the two component model of upstream conserva tive region ribosomal binding site the posi tional nucleotide frequency model and spacer length distribution into the GeneMark hmm algorithm Critical Parameters and Troubleshooting The only program for which the users have the option to adjust the parameters is Gene Mark The main parameter of interest is the Threshold value which determines the level of coding potential above which a r
259. ntact licensing blast wustl edu Repbase Update repeat libraries see Basic Protocol 2 Files A FASTA file or a collection of FASTA files APPENDIX 1B Note that there is no size limit for running RepeatMasker with WU BLAST on command line The example file used in this protocol is the fully sequenced whole C elegans genome 102 287 094 bp in length downloaded from the WormBase http www wormbase org FTP site ftp ftp wormbase org pub wormbase genomes elegans sequences dna 1 Download and install programs RepeatMasker WU BLAST and Repeat library files Note that until June 2004 MaskerAid Bedell et al 2000 was necessary for the WU BLAST to be used with the RepeatMasker That functionality is now implemented and does not need to be integrated separately For this example make a directory called repeat and then copy the RepeatMasker directory into this directory To do this first change to the home directory and then make a new directory named repeat using mkdir Use cd to change directory to repeat as follows mta57 grouse mkdir repeat mta57 grouse cd repeat Current Protocols in Bioinformatics ALTERNATE PROTOCOL Finding Genes 4 10 11 Supplement 25 Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences 4 10 12 Supplement 25 Copy RepeatMasker into this directory Copy WU BLAST package into this directory as well and unpack it mta57 grouse repeat gunzip WU_BL
260. ntamer p and frame j I h Io h LI p log Then given a sequence S of length in frame j the coding potential of the sequence is defined as l 5 LM S LI Sis SOLFI Si i45 i l where S is the subsequence of S starting in position and ending in position j The score of a potential exon S Lg S defined by sites sa start acceptor and sq donor stop is computed as the following log likelihood score Lp S La Sg Lp s4 t Ly S Assembling genes geneid predicts gene structures which can be multiple genes in both strands as sequences of frame compatible nonoverlapping exons If Current Protocols in Bioinformatics a gene structure g is a sequence of exons e1 2 en the score of the gene is the log ratio L g Lp e Le er Le En In geneid the gene structure predicted for a given sequence is the gene maximizing Lc g among all those gene structures that can be as sembled from the set of predicted exons An efficient dynamic programming algorithm is used to find the gene structure G maximiz ing Lg Guig 1998 Actually because of a number of approximations made the simple sum of log likelihood ratios does not produce necessarily genes with the right number of exons if Lg tends to be positive the genes tend to have a large number of exons if Lg tends to be negative the genes tend to have an small number of exons and the score of the ex
261. nus cinereus Triticum aestivum wheat Arabidopsis thaliana Oryza sativa rice Prediction modes Normal mode signal exon and gene prediction Exon mode only signals and exons omitting evidences Assembling mode only assembling evidences DNA strands Forward and Reverse Forward Reverse Figure 4 3 11 geneid Web server Prediction Options area Current Protocols in Bioinformatics Output options t Output format or Elements to be included in the output Signals Exons Figure 4 3 12 geneid Web server Output Options area In the Organism menu users will select the suitable organism depending on the species that the DNA sequence is from see Guidelines for Understanding Results The Param eter File Currently the available organisms are Homo sapiens default Drosophila melanogaster Tetraodon nigroviridis Caenorhabditis elegans Dictyostelium discoideum Plasmodium falciparum Aspergillus nidulans Neurospora crassa Cryptococcus neo morfans Coprinus cinereus Triticum aestivum Arabidopsis thaliana and Oryza sativa In the Prediction Modes menu the geneid engine can be configured to predict either signals exons or genes Depending on the input information users will select Normal mode to obtain the optimal genes predicted on the sequence see Basic Protocol 1 steps 1 and 2 or to reannotate the current sequence by using external informa
262. o consecutive fragments signals found both in forward and reverse strands are displayed for every fragment after a header specifying the fragment positions in the input sequence 4 Obtain the set of predicted First exons along the input sequence by typing geneid P param human3iso param fo samples examplel fa Current Protocols in Bioinformatics geneid can also print all candidate exons along the query sequence The options f i t and s are provided to print the predicted exons of each class First Internal Terminal and Single The options can be combined to print more than one class of exons In such a case exons are printed separately by class If exons of all classes are to be printed it is advisable to use just the option x which prints the list of all exons sorted by position As shown in Figure 4 3 2 bottom each predicted exon is printed in a separate record containing the fields 1 to 11 as described in Table 4 3 1 plus the length and the amino acid sequence of the exon Obtain a more complete output by using the option X geneid P param human3iso param X samples examplel fa By using the option X geneid produces a more exhaustive output of the gene prediction Each exon is now described in three different lines Fig 4 3 3 The first one describes the exon start signal as in step 3 the second line describes the exon itself as in step 4 and the third line describes the exon end signal as in step 3
263. oding RNA genes One such gene NTT noncoding transcript in T cells shows no exons or significant open reading frames even though RT PCR shows that NTT is transcribed as a polyadenylated 17 kb mRNA Liu et al 1997 A similar protein IPW is involved in imprinting and its expression is correlated to the incidence of Prader Willi syndrome Wevrick et al 1996 Since hallmark features of gene structure are presumably absent from such genes they cannot be reliably detected by any method known to date It is becoming evident that no one program provides the foolproof key to computational gene identification The correct choice of program will depend on the nature of the data and where in the pathway of data maturation that data lies Users should always take a combinatorial approach to gene prediction looking for consensus between several meth ods before drawing conclusions about a region of interest consistency among methods can be used as a qualitative measure of the robustness of the results Furthermore use of comparative search methods such as BLAST Altschul et al 1997 UNITS 3 3 amp 3 4 or FASTA Pearson et al 1997 UNIT3 9 should be considered an absolute requirement with users targeting both dbEST and the protein databases for homology based clues A good example of the combinatorial approach is illustrated in the case of the gene for cerebral cavernous malformation CCM located at 7q21 to 7q22 here a combination of MZEF
264. of GlimmerM required some human intervention in the training protocol in particular a programmer or biologist was required to choose thresholds for the false negative and false positive rates for the splice site recognition routines Fortunately the automatic training procedure in the current version obviates this requirement To allow greater flexibility in tuning the system the training procedure permits the user to consult the false positive and false negative rates deter mined from the training data and to adjust the corresponding system parameters The Support Protocol gives the user the necessary knowledge for changing these default thresholds and other parameters of the gene finder Adjustment of these parameters frequently yields better gene predictions because of the wide variation in DNA sequence characteristics for different organisms An Alternate Protocol briefly describes running GlimmerM from the TIGR Web site However individuals who choose to run GlimmerM over the internet do not have the option of training the system RUNNING GlimmerM LOCALLY TO IDENTIFY GENES BASIC The most powerful and flexible way of using GlimmerM is to install and run the PROTOCOL Unix based software on a local system This gives the user more organism specific versions of GlimmerM these are included with the software and the power to train the system for any organism of choice provided that one can collect a representative training set see Support Pro
265. of using the GlimmerM Web server Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 12 Minimum gene length is the length in nucleotides of the smallest fragment permitted to be called a gene Minimum overlap length is a lower bound on the number of bases overlap between two genes that is considered to be a problem Overlaps shorter than this are ignored Minimum overlap percent is another lower bound on overlap Overlaps shorter than this percentage of both genes are ignored Threshold score is the minimum in frame score for a fragment to be considered a potential gene and it is computed as in Salzberg et al 1998a Current Protocols in Bioinformatics File Edt View Go Communicator lt x x a x i 3 e u Back Retoad Home Search Netscape Print Security Shop Bookmarks Netete rtp vw tigr org tigr sccipts glianern glinmern cgi A EJ What s Related y P Tiy Email Z CO Docs Chapter 3 Functions AltaVista Search g Ermez Home g Logon Google Kinima gene length 175 Minimum overlap length 30 Minimum ov percent 10 0 Threshold score 99 Use independent scores True Use first start codon True 6 IndScore Splite sites scores i 0 0 start 0 796904 5 5 0 0 start 0 796904 5 5977 0 start 0 796904 Putative Genes i 273 662 3 Bed overlap with gene 6 2 250 1171 76 00 Bad overlap with gene 5 76 00 86 00 67 00 86 00 68 00 Bad overlap with gene 1234 1371 a
266. om c_p cpbi_sampledatafiles htm 1 Download and install GlimmerM see Basic Protocol 2 Change to the train subdirectory Compile the training module by running make inthe train subdirectory of the package Gl limmerM gt cd train Gl limmerM train gt make 3 From a Unix console or shell window train GlimmerM with the following command trainGlimmerM lt mfasta_file gt lt exon_file gt optional parameters lt mfasta_file gt and lt exon_file gt contain the names of the FASTA file and the file containing the exon coordinates of the known genes respectively A concrete example of running trainGlimmerM for malaria data is presented in Figure 4 4 1 One of the main steps of GlimmerM s gene finding algorithm is determining potential splice sites in the DNA sequences provided as input Splice site sequences that contain the consensus GT or AG dinucleotides and score above a fixed threshold are retained as potential donor or acceptor sites These are filtered further by keeping only those sequences whose score was maximal within a fixed DNA window Pertea et al 2001 The default length of this window is 60 bp but it can be changed by using two optional parameters with the trainGlimmerM procedure a filter value where filter value is an integer specifying the window length for filtering locally maximal acceptor sites default 60 d filter value where filter value is an integer specifying the window
267. on of human gene core promoters in silico Genome Res 8 319 326 Zhang M Q 1998c Statistical features of human exons and their flanking regions Hum Mol Genet 7 919 932 Current Protocols in Bioinformatics Finding Genes 4 2 13 Using MZEF to Find Internal Coding Exons 4 2 14 Zhang M Q 2000 Discriminant analysis and its application in DNA sequence motif recognition Briefings in Bioinformatics 1 331 342 Key References Zhang 1997 See above This is the original MZEF paper Zhang 1998c See above This has human exon classification and feature statistics Zhang 2000 See above This is a tutorial on discriminant analysis and has examples on how to combine MZEF with other programs Contributed by Michael Q Zhang Cold Spring Harbor Laboratory Cold Spring Harbor New York APPENDIX Discriminant Analysis and Bayes Error MZEF is based on a classical discrimination method QDA Quadratic Discriminant Analysis which is a direct descendant of LDA Linear Discriminant Analysis Discrimi nant analysis belongs to general statistical pattern recognition methods and has been widely used in many fields for optimal classification e g Fukunaga 1990 Discriminant analysis is used to answer the following question given N objects how can one assign each object into K known classes with minimum error For simplicity the case of K 2 is only considered although the theory can be easily generalized to K
268. on used in Basic Protocol 1 and 3 1 Input DNA sequence and external information using the Input Data section Fig 4 3 10 This section contains two text areas one to copy paste the DNA sequence and the other for the external information to improve the predictions see Basic Protocol 3 as well as a button to select a graphical representation of the results Users must input a DNA sequence in FASTA format APPENDIX 1B either from file or from the text area while the external information in GFF format is optional The process for building a graphical representation from the geneid output with the program gff2ps can be time consuming when the length of the input sequence is more than 100 kbp and the geneid server might dismiss the query to prevent overloading the server By default this option is disabled 2 Configure Prediction Options Fig 4 3 11 There are three different geneid features to configure the organism the mode and the DNA strands to be scanned for genes All of these fields share the same structure a set of possible values from which the user can only select one Prediction Options t Organism Homo sapiens human Drosophila melanogaster fruit fly Tetraodon nigroviridis puffer fish Caenorhabditis elegans worm Dictyostelium discoideum slime mold Plasmodium falciparum malaria parasite Aspergillus nidulans Neurospora crassa Cryptococcus neomorfans Copri
269. ons is corrected by adding a constant EW Thus given an exon e the actual score of e is Ly e Lg e EW To estimate this constant a simple opti mization procedure is performed The value of EW affects the resulting predictions and it may occasionally be useful to alter its de fault value see Critical Parameters and Trou bleshooting Examples of large scale genomic annotation using geneid geneid has been used in several genome an notation projects as the main gene prediction tool or as a component of the ab initio gene prediction pipeline For instance geneid was the main ab initio gene prediction tool in the Dictyostelium discoideum genome project Glockner et al 2002 geneid has also been used in the large scale analysis of the genome sequence of Tetraodon nigroviridis Jaillon et al 2004 and Paramecium tetraurelia Aury et al 2006 Geneid is currently being used in the annotation of O dioica several fungi the wine grape tomato pea aphid and other genomes geneid predictions in human mouse and other species are served via DAS Distributed Annotation System through the UCSC genome browser http genome ucsc edu and the Ensembl site http www ensembl org They can also be found at the geneid Web site geneid has also been used to scan the Drosophila melanogaster and the Takifugu rubripes genomes for putative selenoproteins Castellano et al 2001 2004 See Sugges tions for Further
270. ons are used for determining what kind of repeats are masked Com monly used options within this category are cutoff nolow and div The option cutoff sets cutoff score for masking repeats when using 1ib The default cutoff score is 225 Lower scores give more false matches A nolow flag causes RepeatMasker not to mask low complexity DNA or simple repeats The div option sets the divergence level to limit the masking and annotation to a subset of less diverged younger repeats c Some options are used to control processing speed and search parameters Options that affect processing speed are q quick search 5 to 10 less sensitive 3x to 4x faster than default qq rush job 10 less sensitive s slow search 0 to 5 more sensitive 2 5x slower than default These flags make significant differences when the input sequences are long If only a quick check is desired the qq flag may be used for fast results On the other hand if the quality of the result is more critical the default with none of the above options selected or even s should be used It is possible to recruit more processors for RepeatMasker by using the pa rallel flag which only works when there are many input files or if the query files are big gt 50 kb WU BLAST can be used to replace cross_match if the flag w ublast see Alternate Protocol is used Current Protocols in Bioinformatics d Output options support the following frequently used
271. onsensus These methods are used to detect features such as donor and acceptor splice sites binding sites for transcription factors poly A tracts and start and stop codons Finally comparative methods make determinations based on sequence homology Here translated sequences are subjected to database searches against protein sequences e g BLASTX unir 3 4 in order to see whether a previously characterized coding region corresponds to a region in the query sequence While this is conceptually the most straightforward of the three approaches it is restrictive in that most newly discovered genes do not have gene products that match anything in the protein databases Also the modular nature of proteins and the fact that there are only a limited number of protein motifs Chothia and Lesk 1986 makes it difficult to predict anything more than just exonic regions in this way The reader is referred to a number of excellent reviews detailing the theoretical underpinnings of these various classes of methods Claverie 1997a b 1998 Guig6 1997 Snyder and Stormo 1997 Stormo 2000 Rogic et al 2001 While many of the gene prediction methods belong strictly to one of these three classes of methods most of those that will be discussed in this chapter combine the strength of different classes of methods in order to optimize their predictions HOW WELL DO THE METHODS WORK Given the complexity of the problem at hand and the range of approaches for ta
272. ormance reported by the developers of the method themselves were used Current Protocols in Bioinformatics in making the comparisons Based on these comparisons the best overall individual exon finder was deemed to be MZEF unir 4 2 and the best gene structure prediction program was deemed to be GENSCAN Back calculating as best as possible from the numbers reported in the Claverie paper these two methods gave the highest correlation coefficients within their class with CCyzpp 0 79 and CCgenscan 0 86 Since these gene finding programs are undergoing constant evolution adding new features and incorporating new biological information the idea of a comparative analysis of a number of representative algorithms was recently revisited Rogic et al 2001 One of the encouraging outcomes of this study was that these newer methods as a whole did a substantially better job in accurately predicting gene structures than their predecessors Using an independent data set containing 195 sequences from GenBank in which intron exon boundaries have been annotated GENSCAN and HMMgene appeared to perform the best both having a correlation coefficient of 0 91 Note the improvement of CCcenscan from the time of the Burset and Guig6 study to the time of the Rogic et al study STRATEGIES AND CONSIDERATIONS Given these statistics it can be concluded that both MZEF unir4 2 and GENSCAN are particularly suited for differentiating introns from exons at di
273. ould collect more known genes with introns and then try the training procedure again If insufficient data is available to train the internal model of translational Start sites the training procedure will succeed but GlimmerM will consider any ATG a potential start site A flag on line 14 of config_ file see step 5 below will indicate if there was enough data to train the start sites this flag is equal to 0 in the case of insufficient data 4 Change to the newly created TrainGlimmM date time subdirectory View the config file Fig 4 4 2 and the TrainGlimmM date time log file Fig 4 4 3 by using a text editor see APPENDIX IC The trainGlimmerM program creates a log file and a subdirectory under the directory where the user ran the training procedure The log file called Train GlimmM date time log Fig 4 4 3 can be consulted to find the default values used for some of the parameters of GlimmerM This subdirectory is called Train GlimmM date time where date and time specify the date and time when the subdirectory was created TrainGlimmM date time contains the training parameters needed by GlimmerM to run The subdirectory also contains a configuration file called config_file Fig 4 4 2 that specifies the parameters in Table 4 4 2 5 Ifnecessary modify the parameters inthe config _ file obtained from the training routines see Critical Parameters and Troubleshooting This step is optional but the authors d
274. ovsky M GeneMarkS a self training method for predicition of gene starts in microbial genomes Imelicatio DS for finding sequence motifs in regulatory regions Nucleic Ac ids Resear cn 2 e005 V ol 29 No 12 2607 2618 Re yd f The models used by GeneMark hmm 2 1 are derived in an iterative manner from the input sequence This program was designed to analyze anonymous prokaryotic genome sized sequences and a5 such we recommend submissions of at least 1 MB Please note that email is the only way to receive output from this web page Here are the GeneMarks edictions for several complete genomes The latest use of GeneMarks is described in Dr RAPSr Option for analyzing eukaryotic viruses has been added formerly GeneMarks Title optional e Sequence Sequence File uplosd e Running Options Use Eukaryotic Virus Version Output Options Email Address required e Generate PostScript graphics Obtain GeneMark 2 4 predictions in addition to GeneMark hmm 2 0 predictions Translate GeneMarkS predicted genes into proteins Figure 4 5 8 The user interface for the GeneMarkS model building program Required input includes a DNA sequence either copied and pasted into the text box or uploaded as a file from the user s computer in FASTA format This program builds a model from the sequence and then runs GeneMark hmm E mail is the only way to receive output from this program symbols of the four nucleotid
275. oximal Check Seq Exon Acceptor Site Donor Site gi 178343 gb M11817 1854 Possibly TRUE Possibly TRUE gi 178343 gb M14076 4208 Possibly TRUE Possibly TRUE gi 178343 gb M16041 6252 Possibly TRUE Possibly TRUE gi 178343 gb M16072 6252 FALSE Possibly TRUE gi 178343 gb M16802 6934 Possibly TRUE Possibly TRUE gil 178343 gb M17759 7856 Possibly TRUE Possibly TRUE gi 178343 gb M19444 9573 Possibly TRUE Possibly TRUE gil178343 gb M19449 9573 FALSE Possibly TRUE gi 178343 gb 0867 11081 Possibly TRUE Possibly TRUE gi 178343 gb M110914 11081 Possibly TRUE Possibly TRUE gi 178343 gb M112481 12613 Possibly TRUE Possibly TRUE gil178343 gb 2505 12613 FALSE Possibly TRUE gi 178343 gb M113341 13425 Possibly TRUE Possibly TRUE gi 178343 gb 3357 13425 Possibly TRUE Possibly TRUE gi 178343 gb M113702 13799 Possibly TRUE Possibly TRUE gil178343 gb M113730 13799 FALSE Possibly TRUE gi 178343 gb M114977 15115 Possibly TRUE Possibly TRUE gil 178343 gb M115534 15757 Possibly TRUE Possibly TRUE gil178343 gb 6941 17073 FALSE Possibly TRUE gil178343 gb M116969 17073 FALSE Possibly TRUE Figure 4 2 5 Prediction results from MZEF SPC In addition to the three user controllable parameters there are also a few hard coded MZEF parameters Minimum ORF size 18 bp because shorter exons are extremely rare Maximum ORF size 999 bp which was chosen according to the longest internal coding exon in the training set Minimu
276. pecies as of August 2002 A particular model is used to analyze a DNA sequence of the same organism The algorithm of the GeneMark program and preliminary studies of the statistical models of DNA sequences have been described in several publications Borodovsky et al 1986a b c Borodovsky and McIninch 1993 The algorithm can determine a posteriori probabilities of protein coding in each of six possible frames for any given DNA sequence fragment The output of the GeneMark program provides these a posteriori probability values as functions of sequence position Then an ORF open reading frame is identified as a gene based on the value of the a posteriori probability it accumulates The text output of the program contains a list of predicted protein coding ORFs Fig 4 5 2 Optionally the a posteriori probabilities for a given sequence can be viewed as a graph in six panels Fig 4 5 3 The default parameters of the program and the format of both the text and graphical outputs can be changed by the user A sample sequence example fna is used below to illustrate how to use the GeneMark Web interface Contributed by Mark Borodovsky Ryan Mills John Besemer and Alex Lomsadze Current Protocols in Bioinformatics 2003 4 5 1 4 5 16 Copyright 2003 by John Wiley amp Sons Inc UNIT 4 5 BASIC PROTOCOL 1 Finding Genes 4 5 1 Supplement 1 I WebGeneNark Microsoft Internet Explorer Ce Gi yos fpotes less ib O n
277. play options are optimized to deal with large genomic regions The UCSC genome browser Hinrichs et al 2006 is a Web tool developed at the University of California Santa Cruz to display on the Internet the available annotations on multiple genomes such as human mouse or fly The UCSC genome browser shows the information in different tracks each one containing different types of genomic information e g genes CpG islands ESTs etc Apart from the annotations provided by the system the users can easily import predictions in GFF format Thus user data and genomic annotations are displayed in the same range of chromosomic coordinates ENSEMBL uniT 1 15 and other genome browsers can also be used in a similar way to display geneid predictions Necessary Resources Hardware gff2ps Unix Linux workstation apollo Unix Windows Macintosh workstation with at least 164 Mb RAM recommended UCSC browser A computer and a connection to the Internet Software gff2ps see Support Protocol for obtaining gff2ps apollo see UNIT 8 5 and Support Protocol for obtaining apollo UCSC Browser Unix Windows text editor See APPENDIX 1C an Internet browser e g Firefox or Internet explorer Files geneid output see Basic Protocol 1 step 6 Visualization using gff2ps la Run gff2ps on the geneid output extracted from Basic Protocol 1 step 6 to obtain a high quality graphical output using the command sgff2ps geneid output gff gt gene
278. prediction programs in a subset of the human genome sequence see Guig6 et al 2006 Although a large fraction of the existing genes will be at least partially predicted by existing tools only a small fraction will be predicted in a completely correct fashion Current Protocols in Bioinformatics On the other hand gene finders tend to overpredict genes resulting in a large number of false positive gene predictions Current methods deal poorly with not so uncommon phenomena such as alternative splicing genes with unusual codon composition nested genes genes within introns noncanonical splice sites and exceptions to the standard genetic code such as those characterizing the selenoproteins Gene boundaries are also poorly predicted often resulting in split or chimeric gene predictions All these drawbacks need to be taken into consideration when interpreting the results of gene prediction programs not only those of geneid see Zhang 2002 for a review The following discusses some more specific features of geneid Accuracy of Geneid Specificity Versus Sensitivity As discussed above most gene finders suffer from lack of specificity predicting a large number of false positive exons and genes particularly in large genomic sequences The authors believe that comparatively geneid has superior specificity to other existing gene finders showing a somewhat more conservative behavior The price is paid in terms of sensitivity geneid v1 2 m
279. ptions are only available for the P falicparum version of GlimmerM SUPPORT PROTOCOL Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 4 TRAINING GlimmerM FOR A SPECIFIC ORGANISM First of all a careful thorough collection of a good training set is a critical first step in the training of any gene finder The quality of the data used for training is directly proportional to the accuracy of the resulting gene finder As with any species specific gene finder GlimmerM needs to learn about the properties of the genes in an organism before it can find more genes A good training set should contain as many complete coding sequences as possible from the organism for which a gene finder is needed It is difficult to specify precisely how many genes are sufficient to form an adequate training set because this number is influenced by several factors such as the length of the ORFs and the number of confirmed splice sites that these genes contain Estimating the parameters of a complex model involving Markov chains like the one used by the authors splice de tection module see Background Information is not an easy task As Burge 1997 shows at least 700 splice site sequences will give a tolerable range of error between 10 and 20 in the estimation of the first order Markov transition probabilities By surveying the public databases one can obtain all previously discovered genes for the target organism and if possible these
280. r which N SCAN has been optimized In all cases the authors of this unit are actively working on enhancing the underlying genome models so significant accuracy improvements are expected in the future Mammals At the time this unit was writ ten N SCAN appears to be the most accu rate de novo gene predictor for mammalian sequences Guigo et al 2006 Stanke et al 2006 In particular it is much more specific than other systems with comparable sensitivity Gross and Brent 2006 Guigo et al 2006 Further N SCAN is particularly good at the extremely challenging problem of exact gene prediction relative to other systems to which it has been compared Nucleotide sensitivity is in the 90 to 95 range exact exon sensitiv ity in the 75 to 85 range and exact gene sensitivity in the 30 to 40 range Speci ficities are probably in the same ranges It is easier to measure sensitivity than specificity because sensitivity can be calculated on a par tially annotated genome In contrast all genes should be known to measure specificity cor rectly and currently no genome is annotated entirely Therefore the specificity measure is always an underestimation C elegans Current gene finders are more accurate on Caenorhabditis genomes than on mammals by roughly 5 to 10 on the nu cleotide and exon levels and 25 to 30 on the exact gene level TWINSCAN the prede cessor of N SCAN appears to be more accu Current Protocols in Bio
281. rand that fall inside a gene transcript assembled transcript of cCDNA EST are potential false positives and should be ignored COMMENTARY Background Information First exons promoters and CpG windows Gene finding is one of the most vital phases of genome annotation Sequence homology is perhaps the most important evidence used to detect functional elements in genomic se quences A direct comparison of a genomic sequence Example 1 with cDNA ESTs can identify regions of the query sequence that correspond to transcribed genes However most of the ESTs and cDNA sequences are 5 incomplete and do not provide information about the first exons and promoter regions On the other hand gene finding programs such as GENSCAN Burge and Karlin 1997 and MZEF Zhang 1997 unr 4 2 were trained to predict protein coding exons only Detecting the gene regulatory regions is important not only to annotate the genome but also help to understand the large scale gene expression data such as those from microarray experi ments see Chapter 7 FirstEF is the only pro gram that was specifically designed to predict both partially coding and noncoding first ex ons In the human genome 40 of the genes have completely noncoding first exons and first introns tend to be longer than average Davuluri et al 2001 Stretches of DNA gt 200 nucleotides with high G C content and a frequency of CpG dinucleotides close to the expected value are
282. re returned in the browser window Necessary Resources Hardware Any computer workstation PC Macintosh Unix Linux with Web access Software Web browser e g Netscape Navigator Microsoft Internet Explorer Files DNA sequence of interest in Raw or FASTA format APPENDIX 1B 1 Open the GrailEXP Web page http compbio ornl gov grailexp in the Web browser The GrailEXP Web page will appear in the main browser window Fig 4 9 1 The sidebar on the left side of the page provides useful links including references the FAQ Frequently Asked Questions and license agreement for downloading the software The main frame on the page displays the analysis request submission form which provides for user selection of options and sequence input for the analysis as outlined in the following steps Contributed by Edward C Uberbacher Doug Hyatt and Manesh Shah Current Protocols in Bioinformatics 2003 4 9 1 4 9 15 Copyright 2004 by John Wiley amp Sons Inc UNIT 4 9 BASIC PROTOCOL Finding Genes 4 9 1 Supplement 8 Z Grail ExP Home Page Microsoft Internet Explorer A File Edit View Favorites Tools Help esk gt OA A Asearch fqravorites PMedia A B S GY S Address http compbio arnl govigrailexpy x so Links Home FAQ News Download References es Of Google v Psearch Web GiSearchsite F3news EPrage info fup Hohioht Norton Antiviru
283. rea For the purpose of analysis all nonalphabet symbols are ignored and all ambiguous letters other than the symbols of the four nucleotides assuming that they occur rarely are replaced with C This minimizes the chance of the possible creation of a false start or stop codon 2 Scroll down the page and set the Running Options Select the name of the species of interest from the Species pull down menu which will result in the selection of the corresponding statistical model The other pull down menus Window Size Step Size and Threshold are set at default values chosen for being optimal in average results however the user has the option to change the default values RBS models are available for some species Models are available for some species with an alternative genetic code Choosing the correct species name is essential since wrong statistical models may totally corrupt the results of gene prediction Sequences of species for which no model is available should be analyzed using either the heuristic method see Basic Protocol 2 or GeneMarkS see Alternate Protocol 2 3 Scroll further down the page and select the Output Options By default the program generates text output in the form of a list of open reading frames predicted as coding sequences The optional graphical output will be sent to the E mail address provided by the user 4 After completing the above entries click the Start GeneMark button The results will be depicte
284. reading frame scores above the score set by the t option by default t 90 then that model is predicted to be a gene If the x option is added when running GlimmerM then the score of the putative coding region in the correct reading frame is also compared to the score generated by a random model which is a simple Markov chain that uses independent probabilities for each base See for example Salzberg etal 1999 for a description of how to use Markov chains for biological sequence analysis The r option is active by default but it can be disabled by adding r to the command line By default GlimmerM uses a maximal local filtering for the splice sites with a window length read from the config_file This is equivalent to using the option when running GlimmerM Because the filter may increase the number of false negatives the option should be used when no filtering is desired The splice site thresholds that the program reads from config _filecanalso be changed with the 5 and 3 parameters in this way overriding the initial threshold values given in the config_file If enough data is available GlimmerM will train a module to reduce the false positive rates of the translational start recognition If the user does not wish to use this module the s option should be used When the s parameter is not specified the s option is enabled if possible 4 Examine the results see Guidelines for Understanding Results The res
285. rent Protocols in Bioinformatics Finding Genes 4 3 7 Supplement 18 Using geneid to Identify Genes 4 3 8 Supplement 18 example1_1_ Figure 4 3 5 Using gff2ps to visualize geneid output Graphical representation of geneid output on sequence example1 with default gff2ps Visualization using apollo 1b To start an apollo session type SApollo The load data window will appear 2b Choose Ensembl GFF file format as a data source Select a file for the visualization by writing the entire path in the GFF file box or by browsing the directory tree Select geneid_output gff then click the OK button The main apollo window will appear Fig 4 3 6 shows the default apollo display of the prediction obtained in Basic Protocol 1 step 6 Coding exons provided by geneid are displayed below the main toolbar Exons predicted on the forward strand are displayed File Edit View Tiers Analysis Bookmarks Annotation Window Links Help Position 4 WHHL Zoom x10 x2 x5 x1 Reset Zoom factor 1 0000 Range qeneid_yv1 2 examplel 736 31947 6 14 lexamplel Genomic Range Genomic Length Score 736 1130 395 5504 5618 115 5778 5951 174 8730 8836 107 13186 13256 71 21287 21488 202 29896 30019 124 31726 31947 222 Position 23979 Feature Figure 4 3 6 Using Apollo to visualize geneid output Current Protocols in Bioinformatics a
286. results both private and published Users of the Web server cannot choose among parameter sets Users with local installations can choose input pa rameter files from the N SCAN package or provide their own To estimate parameters the program iPEstimate can be downloaded from http mblab wustl edu software iPEstimate is a versatile parameter estimation program and comes with detailed information on both pa rameter estimation theory and use of the program For collaboration on adapting N SCAN to new genomes E mail nscan mblab wustl edu Masking Masking of simple and interspersed repeats is also a consideration In general N SCAN is both faster and more accurate on sequences that have been aggressively masked for inter Current Protocols in Bioinformatics spersed repeat elements However masking of simple and low complexity repeats results in slightly lower sensitivity because some gen uine exons do contain such repeats The au thors have found that N SCAN is most ac curate when interspersed repeats are masked and low complexity simple repeats are not therefore these are the default settings How ever this may lead to overprediction of repeat containing exons and in this case it is advis able to try another round of prediction with the simple and low complexity repeats masked Suggestions for Further Analysis The N SCAN group has developed a soft ware package called Eval Keibler and Brent 2003 for compa
287. results page In addition check the program Current Protocols in Bioinformatics SUPPORT PROTOCOL Finding Genes 4 8 11 Supplement 20 Using N SCAN or TWINSCAN 4 8 12 Supplement 20 logs under the Logs button to see if any of the procedures returned an error Questions and comments regarding unexpected results may be sent to nscan mblab wustl edu in a message that includes the job I D Do not forget that unexpected results are sometimes correct COMMENTARY Background Information Approaches to gene structure prediction Currently there are two major approaches to automated gene prediction De novo sys tems take one or more genomes as input while expression based systems use databases of known transcripts to annotate a genome Single genome ab initio or de novo gene predictors take only the target genome as in put and a probabilistic model that abstracts patterns common to many genes is used to an notate the genome Examples of single de novo gene predictors are GENSCAN Burge 1997 Burge and Karlin 1997 FGeneSH Salamov and Solovyev 2000 and Augustus Stanke and Waack 2003 Stanke et al 2006 A sec ond approach to de novo gene prediction typ ified by TWINSCAN N SCAN Korf et al 2001 Flicek et al 2003 Gross and Brent 2006 SLAM Alexandersson et al 2003 and SGP2 Parra et al 2003 augments a de novo system with information based on alignments between the target genome and one
288. rg cgi bin WEBRepeatMasker 2 Create an alignment file using the informant sequences The informant alignment file consists of a FASTA header line and a line for each informant The length of each informant line is the same as the length of the target sequence For each character in the target sequence there is a corresponding character in each informant sequence The informant sequence alphabet is A C G T where an informant character from the set A C G T means the informant character aligns to the corresponding target characters _ is used for informant gaps within aligned regions and is used for target regions to which the given informant does not align To create the alignment sequence for a single informant run Blastz and convert its output using the following commands the greater than sign before the output file name is interpreted by Unix as specifying where the output should be stored blastz masked target sequence informant sequence gt target lav lav2maf target lav masked target sequence informant sequence gt target maf maf_to_align pl output directory target maf ascending masked target sequence informant sequence gt target align The program maf_to_align pl is included in the N SCAN download package For multiple informants maf files can be generated using multiz For running multiz follow instructions from http www bx psu edu miller _lab In addition multiz alignments in ma
289. rg et al 1999 See above This paper introduces the GlimmerM method in itially used in finding genes in Plasmodium falci parum This paper also describes how GlimmerM was used in the annotation of chromosome 2 of P falciparum Internet Resources http www tigr org software glimmerm GlimmerM Web site http www tigr org tdb edb2 pfa1 htmls A preliminary annotation of chromosomes 10 11 and 14 of P falciparum This will change when the P falciparum genome is completed Contributed by Mihaela Pertea and Steven L Salzberg The Institute for Genomic Research Rockville Maryland Current Protocols in Bioinformatics Prokaryotic Gene Prediction Using GeneMark and GeneMark hmm In this unit the GeneMark and GeneMark hmm programs are presented as two different methods for the in silico prediction of genes in prokaryotes GeneMark see Basic Protocol 1 which uses Markov chain models and Bayes rule Durbin et al 1998 to predict protein coding and noncoding regions can be used for whole genome analysis as well as for the local analysis of a particular gene and its surrounding regions Gene Mark hmm see Alternate Protocol 1 makes use of hidden Markov models to find the transition points boundaries between protein coding states and noncoding states and can be efficiently used for larger genome sequences These methods can be used in conjunction with each other for a higher sensitivity of gene detection They both
290. ring predictions generated by different programs with each other or with standard annotations It provides sum maries and graphical distributions for many statistics describing any set of annotations regardless of their source It also com pares sets of predictions to standard anno tations and to one another Eval is open source software and can be obtained from http mblab wustl edulsoftware Acknowledgements Thanks to Randall Brown for unpublished data on the effects of evolutionary distance on N SCAN accuracy to Laura Langton for error proofing the support protocols and to Laura Kyro for creating the figures for this unit This work was supported in part by grants from the National Institutes of Health HG002278 HG003700 the National Science Foundation 0501758 and Monsanto to Michael R Brent Literature Cited Alexandersson M Cawley S and Pachter L 2003 SLAM Cross species gene finding and alignment with a generalized pair hidden Markov model Genome Res 13 496 502 Allen J E and Salzberg S L 2005 JIGSAW Inte gration of multiple sources of evidence for gene prediction Bioinformatics 21 3596 3603 Allen J E Pertea M and Salzberg S L 2004 Computational gene prediction using multiple sources of evidence Genome Res 14 142 148 Brown R H Gross S S and Brent M R 2005 Begin at the beginning Predicting genes with 5 UTRs Genome Res 15 742 747 Burge C 1997 Identification of
291. rkS in the same manner as for prokaryotes Necessary Resources Hardware A personal computer or workstation with Web access Software A Web browser Files A single sequence in FASTA format APPENDIX 1B The sample sequence example fna which contains region 1 to 50 000 from Escherichia coli K12 used to illustrate this protocol can be downloaded from the Current Protocols Web site http www3 interscience wiley com c_p cpbi_sample datafiles htm 1 Viaa Web browser connect to http opal biology gatech edu GeneMark genemarks cgi In the Input Sequence section paste an input sequence into the Sequence box area or alternatively click on Browse next to the Sequence File Upload box to upload the input sequence file from a local drive The Sequence File Upload option is more powerful since the copy and paste method imposes a limit on the length of the sequence If the sequence has a FASTA APPENDIX 1B title line e g gt Sequence name this name will be assigned to the sequence in the output unless the user gives a name in the Sequence Title text area For the purpose of analysis all nonalphabet symbols are ignored and all ambiguous letters other than the Current Protocols in Bioinformatics J Genebarks Microsott Internet Explorer fie 0a ew fpo ooh tee a A Om O i AG Amn Froes Grete O PD jopa taig gaech Aaret igeremarbs 69 Google CP Sewn wed Esera ste Reference Besemer Lomsadze A and Borod
292. rnal 5778 5951 1 13 00 1 43 0 86 13 14 0 00 AA 171 228 examplei_1 Donor 5951 5952 0 86 AAGGTATAG Acceptor 8729 8730 4 75 0 00 0 00 TITTTGTITTATATTGITTTACAGTGT Internal 8730 8836 0 84 02 4 75 3 34 3 72 0 00 AA 229 264 examplei_1i Donor 8836 8837 3 34 AAGGTATGG Acceptor 13185 13186 1 21 0 00 0 00 GGTTTATTACATTTTTATACATAGAAT Internal 13186 13256 0 46 11 1 21 5 38 5 00 0 00 AA 264 288 examplei_i Donor 13256 13257 5 38 CAGGTAAGA Acceptor 21286 21287 2 27 0 00 0 00 AAAGGTTATITTTATTCAATAAAGTGA Internal 21287 21488 2 78 22 2 27 5 44 15 95 0 00 AA 288 355 examplei_i Donor 21488 21489 5 44 CAGGT AAGC Acceptor 29895 29896 1 27 0 00 0 00 CATGAAAATTAAATTTICTICTAGIGA Internal 29896 30019 1 56 10 1 27 0 56 14 90 0 00 AA 355 396 examplei_1 Donor 30019 30020 0 56 GAGGTATTT Acceptor 31725 31726 1 78 0 00 0 00 TAAAAGTTCCCTTTGTITACTTAGCTT Terminal 31726 31947 3 30 00 1 78 0 00 19 32 0 00 AA 397 470 examplei_1i Stop 31945 31947 0 00 TTAA gt example1_i geneid_v1 2_predicted_protein_11470_AA MGTSGDHDDSFMKMLRSKMGK CCRHCF PCCRGSGT SN VGT SG DHENSF MK MLR SKMGKWC CHCFPCCRGSGKSN VG AWGDY DHS AFMEPR YH IRREDLDK LHRAAWWGKV PRKDLI VMLR DTDMNKRDKEKRTALHLAS ANGNSEVV QLLLDRRCQLNVLDNKKRTAL IK AIQCQEDECV LMLLEHGADRNI PDEYGNT ALHYA TYNEDKLMAKALLLYGAD IESKNKCGLTPLLLGVHE QKQQVVKFLIKKKANLNVLDRYGR ICELLSDY KEK QMLKI SSENSNPV IT ILN IKLPLKV EEE IXKHGSNPVGLPENLTNGASAGNGDDGLI PQRRSRKPENQOF PDTENEEY HSDEQND TRKQLSEEQNTG ISQDEILTNKQKQIEVAEQKMNSELSLSHK
293. rom the user client to the geneid server by clicking on the button Submit Depending on the complexity of the query and the length of the input sequence the results Fig 4 3 13 will be returned to the user in a reasonably short period of time The form can be reset and its content deleted with the button Reset form Users can obtain help through several links in the Web page Necessary Resources Hardware A computer and a connection to the Internet geneid 1 2 Web Server 2005 Paste your FASTA sequence here Or search a FASTA file to process romeseblanco cPByne Browse Paste your GFF evidences here Field separator tab Or search a GFF file containing evidences to process Browse F Do you want a graphical representation of the predictions it might be time consuming depending on the size of the sequence Maximum sequence size for plots 100 000 bps Figure 4 3 10 geneid Web server DNA and external information area Current Protocols in Bioinformatics ALTERNATE PROTOCOL Finding Genes 4 3 15 Supplement 18 Using geneid to Identify Genes 4 3 16 Supplement 18 Software An up to date Internet browser such as Internet Explorer hitp www microsoft comlie Netscape http browser netscape com Firefox http www mozilla org firefox or Safari http www apple com safari Files All of the sequences in FASTA format APPENDIX 1B and external informati
294. rotocol 3 VISUALIZING geneid PREDICTIONS geneid does not produce a graphical output by itself However because it is capable of producing GFF output several GFF visualization tools can be very easily used to display geneid predictions This protocol describes how to use three different tools the programs gff2ps and apollo unr 9 5 and the UCSC genome browser UNIT 1 4 to visualize the output produced by geneid gff2ps Abril and Guig6 2000 was developed at the Institut Municipal d Investigacio Medica IMIM in Barcelona as a visualization tool for genomic sequence annotations gff2ps takes the annotated features on a genomic sequence in GFF format as input and produces a high quality PostScript file It can be used in a very simple way because gff2ps assumes that the GFF file itself carries enough formatting information although it also allows through a number of options and Current Protocols in Bioinformatics aconfiguration file a great degree of customization apollo UN T 9 5 Lewis et al 2002 is a genomic annotation viewer and editor It has been developed as a collaboration between the Berkeley Drosophila Genome Project and The Sanger Centre Cambridge U K apollo has been designed to be a complete genome annotation tool for use as a graphical front end to a database that stores sequence annotations Its interactive interface allows the user to browse among predictions in a very intuitive way Zoomable and scrollable dis
295. s a report of GeneMark predictions by checking Print GeneMark 2 4 Predic tions or additional list of the protein translations of the predicted genes by checking Translate Predicted Genes into Proteins A valid E mail address is required if a sequence is longer than 100 kb or if the graphical output is requested The PostScript file will be returned to the user by E mail 4 After completing the above entries click the Start GeneMark hmm button to start running the program For sequences shorter than 100 kb the result will be displayed on the screen If the user supplied an E mail address and checked Generate PostScript Graphics a PostScript file will be E mailed to the user as well For longer sequences the results can be only obtained via E mail 5 Interpret the text output The eukaryotic GeneMark hmm text output Fig 4 6 2 contains a list of Predicted genes exons for each predicted gene in terms of sequence coordinates Both complete and partial genes are predicted by the program Partial exons are not predicted For each gene identified by the gene number in the first column there could be one or more exons listed on separate lines and identified by the exon number in the second column In the third column exons predicted on the direct strand are indicated with a sign while those predicted on the reverse strand are labeled with a minus sign For genes in which multiple exons were predicted the Exon Type
296. s Ea Credits Acknowledgments Bugs Questions Comments GrailEXP is a software package that predicts exons genes promoters polyas CpG islands EST similarities and repetitive elements within DNA sequence GrailEXP lis used by the Computational Biosciences Section at Oak Ridge National Laboratory to annotate the entire known portion of the human genome including both finished and draft data Ifyou are interested in microbial genome analysis and annotation you should go to the Generation home page Perform Analysis Select organism Human Homo sapiens z Select output type Human Readable Text z 7 Perceval Exon Candidates Locate Grail exons using an improved version ofthe Grail 3 neural net I Galahad EST mRNA cDNA Alignments Search from the selected EST mRNA databases and build exons based on similarities with the sequences in these databases GrailEXP Database Refseq HTDB dbEST EGAD Riken Ey NCBI Refseq mRNAs NCI Mammalian Gene Collection Human NCI Mammalian Gene Collection Mouse Baylor Human Transcript Database TIGR EGAD Transcript Database Riken Fantom Mouse cDNA Database Select database s to search CBIL UPenn DOTS EST Assemblies z C Gawain Gene Models Assemble complete gene structures from the above selected options i e Perceval exon candidates and or Galahad EST mRNA alignments Gene modeling organism options Use ESTs mRNAs from any organism 7 I
297. s able to analyze chromosome size sequences in a few minutes on a standard workstation and has arich set of output options which allow for a detailed analysis of gene features in genomic sequences Both a Web server interface and a stand alone distribution are available This unit describes how to use the geneid Unix application to predict genes along genomic sequences see Basic Protocol 1 These can be multiple genes on both strands of large genome sequences or partial genes or exon signals in small genomic fragments Basic Protocol 1 describes the default behavior of geneid and introduces the basic options for configuring its output Next options for visualizing the output are described see Basic Protocol 2 A third protocol describes how to use geneid together with experimental ev idence or evidence coming from other sources to reannotate sequences whose genomic features have been partially annotated see Basic Protocol 3 Use of the Web server version of geneid is described in the Alternate Protocol The Support Protocol describes how to download the geneid software which is in the public domain under a GNU GPL license http www gnu org Complete up to date documentation is provided with the geneid distribution and can also be accessed through the geneid Web page see Support Protocol USING THE geneid UNIX APPLICATION TO PREDICT GENES geneid can be used in two different ways via a Web server see Alternate Protocol or as a
298. s component On completion of the entire pipeline processing the final pipeline status page is displayed with a button labeled Get Summary On clicking this button the server returns a summary page Fig 4 9 3 with a page similar to the status page but with additional information about the number of hits found by each analysis component and hyperlinks to the individual analysis results accessible in several Current Protocols in Bioinformatics ORNL Genome Analysis Pipeline Eukaryotic Microsoft Internet Explorer aloj xJ Fle EGR View Favorkes Toos Hep el Bak gt O Al Geach raoe Gres G D oI sl Address htto icompbio oni gow igenomepipeline ieuk shtml gt Lbs Googie Gsewch web Geseschsee glvews Pops info a norton Artis 5 Computational Biology at ORNL Analysis Tools Channel Generation Grail e GrailExpP Pipeline Parser PROSPECT ORNL Genome Analysis Pipeline Eukaryotic Select organism Human Select all services l GrailEXP Genes gt Post processing on predicted genes Blastp Database Swssprot E value fie 30 Pfam F F Genscan Genes After Repeat Masking gt Post processing on predicted genes Blastp l Database Swissprot E value 10 30 Pfam f F CpG F RepeatMasker M RNA T BAC end pairs l STS eper M BLASTN Database rt z Options Expect 0 000 z Word Sire 11 gt Hits amp Abgnments 10 DNA Sequence Browse Or Demo Submit Request
299. s corresponding to the exons of the predicted gene the amino acid sequence of the gene is printed in FASTA format APPENDIX 1B The frame and remainder see Table 4 3 1 of an exon are the number of hanging nu cleotides not included in complete codons at the left right ends of exons when these are assembled into a gene The formal definition of geneid frame is The number of nu cleotides 0 1 2 from the first nucleotide in the exon to the first nucleotide in the first complete codon in the same exon The remainder is defined in geneid as The number of nucleotides left 0 1 2 after the last complete codon has been translated from the exon sequence given its frame By definition then all First exons have frame 0 as in Fig 4 3 2 and all Terminal exons have remainder 0 3 Obtain the set of predicted Start codons along the input sequence by typing geneid P param human3iso param bo samples examplel fa In addition to the predicted genes geneid provides a number of options which allow the investigator to print an exhaustive list of all the sequence signals and exons predicted along the query sequence most of which are not included in the final gene prediction This option can be useful for instance to carry out a detailed analysis of a small genomic region for potential alternative splice sites If only information on these sites and exons is required it may be advisable to use the option 0 which switches off the gene ass
300. same form or context The reader is referred to the text for greater detail Current Protocols in Bioinformatics promoter Similarly during end modification the poly A tail may be present or absent or may not contain the canonical AATAAA Adding to these complications is the fact that an open reading frame is required but is not sufficient to identify a region as an exon Given these and other considerations there is at present no straightforward method that will allow 100 confidence in the prediction of introns or exons It is true however that the availability of finished genome sequences from a variety of organisms has led to a better understanding of gene structure and is in turn leading to the development of better methods for identifying genes in what is essentially anonymous DNA CATEGORIZING THE METHODS Briefly gene finding strategies can be grouped into three major categories Content based methods rely on the overall bulk properties of a sequence in making their determinations Characteristics considered here include the frequency at which particular codons are used the periodicity of repeats and the compositional complexity of the sequence Because different organisms use synonymous codons with different frequency such clues can provide insight to help determine which regions are more likely to be exons In site based methods by contrast the focus is on the presence or absence of a specific sequence pattern or c
301. sequence In open ended mode it is common to find single exon partial genes predicted near the edges of the query sequence Single strand gene modeling By default Gawain runs in double stranded gene modeling mode However this can cause problems with recognition of embedded genes which may be omitted or the embedding gene may be broken Gawain can be run on just the forward strand or reverse strand In fact the most accurate way to perform gene modeling with Gawain would be to predict exons alignments in double stranded mode then run the gene assembly program twice once on the forward strand once on the reverse strand ESTs and cDNAs not listed as gene evidence If an EST cDNA is not listed as gene evidence but it is in the same location as the gene model then there is something about that EST cDNA that made it inconsistent with the gene model Either it did not align to the same splice sites or it trickled over the edges into a predicted intron This happens in instances where an EST s first base falls very near a splice site edge often there are not enough bases to align with the next exon upstream so the program winds up extending the alignment into an intron This problem can be solved by an examination of the alignments in the gene modeling phase and subsequent correction of such cases EST cDNAs that are inconsistent with gene models are flagged but are not currently returned to the user If deemed useful a future version of Gawa
302. sequencing and analysis of the human genome Nature 409 860 921 TIoshikhes I and Zhang M Q 2000 Large scale human promoter mapping using CpG islands discrimination Nature Genet 26 61 63 Minghetti P P Ruffner D E Kuang W J Dennison O E Hawkins J W Beattie W G and Dugaiczyk A 1986 Molecular structure of the human albumin gene is revealed by nucleotide sequence within q1 1 22 of chromosome 4 J Biol Chem 261 6747 6757 Modrek B and Lee C A 2002 A genomic view of alternative splicing Nat Genet 30 13 19 Solovyev V V Salamov A A and Lawrence C B 1994 Predicting internal exons by oligonucleotide composition and discriminant analysis of spliceable open reading frames Nucl Acids Res 22 5156 5163 Tabaska J E and Zhang M Q 1999 Detection of polyadenylation signals in human DNA sequences Gene 231 77 86 Tabaska J E Davuluri R and Zhang M Q 2001 A novel 3 terminal exon recognition algorithm Bioinformatics 17 602 607 Thanaraj T A and Robinson A J 2000 Prediction of exact boundaries of exons Briefings in Bioinformatics 1 34356 Zhang M Q 1997 Identification of protein coding regions in the human genome by quadratic discriminant analysis Proc Natl Acad Sci U S A 94 565 568 Zhang M Q 1998a Identification of protein coding regions in Arabidopsis thaliana genome based on quadratic discriminant analysis Plant Mol Biol 37 803 806 Zhang M Q 1998b Identificati
303. service is in a testing phase Please report problems and offer suggestions to John Besemer G Je Length Tue Bar Gone Gene Prediction Results Infermanon on mut seques e Sequence title gt Escherichia coll Fil Region Gec percentage 2 29 t Parse predicted by GeneMark ham 2 1 Genefierk tew PROEARTOTIC Version 2 1 Bodel organismo Preeudonet tve modet Predicted genes New models included for many newly sequenced prokaryotic genomes 0000 tp 32 26 99 2008 Strand 8 LeftEad Righttsa lt 3 bad 190 ass 337 2799 2801 373 3704 soro 343 590 tes oe 6n9 7911 ere 9 93 FIO6 909 one 20404 3O93 1 9356 12982 1706 12163 14079 34368 15390 154145 16557 16300 16720 16751 16999 109 10655 18723 19620 1981 20314 Figure 4 5 5 The text output from the GeneMark hmm program for the example sequence using both the Atypical and Typical model The predicted gene coordinates are listed along with the strand on which they were predicted as well as the model class 1 for Typical and 2 for Atypical from which it was predicted The Sequence File Upload option is more powerful since the copy and paste method imposes a limit on the length of the sequence If the sequence has a FASTA APPENDIX 1B title line e g gt Sequence name this name will be assigned to the sequence in the output unless the user gives a name in the Sequence Title text area For the purpose of analysis all nonalphabet sym
304. ss_match is one of the best applica tions for sequence alignment The drawback of cross_match is that it is slow To make RepeatMasker process faster WU BLAST can be used to replace cross_match see Alternate Protocol The alignment program WU BLAST is a heuristic alignment algo rithm However the sensitivity is reduced How RepeatMasker works RepeatMasker finds and masks repetitive elements by aligning each of the query se quence s with each of the repeat consen sus sequences in the repeat library file Usu ally crossmatch is the engine that does the alignment while RepeatMasker manages the whole process and parses the align ments The program cross_match implements Current Protocols in Bioinformatics Finding Genes 4 10 13 Supplement 25 Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences 4 10 14 Supplement 25 when running RepeatMasker with WU BLAST Critical Parameters and Troubleshooting Limitations and known bugs For files with multiple long sequences e g a file containing whole chromosome se quences RepeatMasker does not work well All of the output entries are mislabeled as the first sequence chromosome There is a de fault maximal sequence length of 4 Mb There are two ways to work around this limitation One way is to change the default maximal sequence length value in the RepeatMasker script Find the following line in the script Smaxsize 4000000
305. ssful it creates a directory called N SCAN 3 Test the installation by changing to the N SCAN directory and entering the following command at the prompt test executable If this test is successful N SCAN can be run as described in Alternate Protocol 1 Otherwise a message will appear indicating that the machine is not compatible with the N SCAN executable and offering further suggestions GUIDELINES FOR UNDERSTANDING RESULTS Analyzing Problems and Errors If the Web server finds a problem with the input sequence it will generally provide an immediate explanation of the problem Mammalian sequences may take 2 to 4 hr to run during which there is no response Heavy traffic on the server may cause any job to wait many hours before running If there is no response 24 hr after submission write to nscan mblab wustl edu and include the submission I D for the job in question If unexpected results are received such as gene density that is much higher or lower than expected the first step is to check that the correct sequence was submitted One useful check is to compare the length of the sequence N SCAN processed which is returned at the top of the Submission results page with the length of the sequence you intended to submit If the sequence is correct verify that the correct target organism was selected and that the masking settings were as intended This information can be found under the Submission Details button on the Submission
306. ssful for finding genes in both prokaryotes and eukaryotes Burge and Karlin 1997 Salzberg et al 1998a To score a se quence using a Markov chain a gene finder needs to compute a set of probabilities from training data These probabilities take the form P b Ib _1 b 2 b 3 where b indicates the base in position i of the DNA sequence A 5 order Markov chain for instance would compute probabilities for each of the 4 bases following every possible 5 base combination i e it would compute 4096 probabilities As with many other gene finders Salzberg et al 1998a there are a number of assump tions used by GlimmerM to simplify the task Finding Genes 4 4 15 Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 16 of gene prediction and narrow the possible choices when making predictions The main assumptions are 1 the coding region of every gene begins with a start codon ATG 2 a gene has no in frame stop codons and no frameshift mutations 3 each exon is in a consistent reading frame with the previous exon and 4 every intron begins and ends with the consensus dinucleotides GT AG These constraints sig nificantly enhance the efficiency of the algo rithm for searching through all possible gene models by restricting the search space of the dynamic programming algorithm On the other hand genuine frame shifts cannot be detected by the system Detecting splice sites To detect splice si
307. ssume however that these isoforms are unknown Since a number of ESTs align to this genomic region supporting alternative 3 end exonic structures this example will show how geneid can be used to extend these EST alignments to recover the full alternative transcript in each case 4 Use geneid to extend a gene structure derived from a given EST Type geneid P param human3iso param R samples example3 EST1 gff samples example3 fa The genomic coordinates of the alignment of one of these ESTs EST1 to the genomic sequence obtained e g using ESTgenome Mott 1997 GeneWise Birney and Durbin 2000 or any other cDNA to genomic DNA alignment tool are included in a GFF file which is passed via the R option into geneid These programs obtain a so called spliced alignment between the EST sequence and the genomic query In such an alignment big gaps iikely to correspond to introns are only allowed at legal splice junctions The GFF file in this case is example3 EST1 Internal 27330 27588 example3 EST1 Internal 28652 28704 example3 EST1 Terminal 29345 30124 The result of the prediction appears in Figure 4 3 9B geneid predicts a product distinct from the default prediction which incorporates the three exons in the EST sequence and which resembles closely one of the known alternative forms for this gene 5 Use geneid to obtain an alternative structure supported by a different EST geneid P param human3iso par
308. st likely donor splice site The resulting alignments each consisting of exons introns and splice sites are the final gene message alignments The program has an effective mechanism to deal with repetitive DNA elements During the initial BLAST search unir 3 3 the program looks for the word repetitive in the FASTA header APPENDIX 1B of the sequence that is hit It also identifies clusters of single alignments in the same location in the sequence that hit por tions of many ESTs If ESTs in this cluster are labeled repetitive it eliminates those align ments No EST is ever eliminated that hits in two or more places in what looks like a potential gene In addition any single fragment align ments that do not cover at least 50 of that EST are eliminated The gene message database can often be quite large as of the time of this writing it is 10 9 Gb In order to perform the database alignment phase with limited resources and to speed up the process Galahad has been de signed to run in serial or parallel mode using a multiple partition version of the database In serial mode each partition is searched sequen tially In parallel mode all partitions are searched concurrently on different machines There are several advantages to the multiple partition scheme over running BLAST mul tithreaded against a huge database The data bases are loaded into memory in parallel Each compute node uses substantially less memory
309. stop sites and splice sites as well as nucleotide frequency tables and length distributions of exons introns and intergenic regions to significantly bolster the accuracy of its predictions Most gene identification programs share several major drawbacks of which users need to be keenly aware Since most of these methods are trained on test data they will work best in finding genes most similar to those in the training sets that is they will work best on things similar to what they have seen before Often methods have an absolute requirement to predict both a discrete beginning and an end to a gene meaning that these methods may miscall a region that consists of either a partial gene or multiple genes The importance given to each individual factor in deciding whether a stretch of sequence is an intron or an exon can also influence outcomes as the weighing of each criterion may be either biased or incorrect One of the methods discussed in this chapter GlimmerM unit 4 4 was originally developed for small eukaryotes with a relatively high gene density but the authors have made it possible for users to train the method on their own set of test data this allows the method to be adapted to the peculiarities of any given organism and circumvents some of the problems involved in using a method optimized for one organism on another Finally there is the unusual case of genes that are transcribed but not translated so called nonc
310. structure Often the results from these programs can reinforce each other For example one could run CorePromoter Zhang 1998b CpG_Promoter Ioshikhes and Zhang 2000 FirstEF a first exon finder Davuluri et al 2001 JTEF a last exon finder Tabaska et al 2001 and Polyadq a polyA site finder Tabaska and Zhang 1999 All these programs can be accessed from http www cshl org mzhanglab Examples of how one can combine some of these programs for gene finding may be found in Zhang 2000 Internet Resources http www cshl org genefinder MZEF Web server http www cshl org mzhanglab Papers and other related information for MZEF ftp cshl org pub science mzhanglab FTP site for MZEF Literature Cited Bishop C M 1996 Neural Networks for Pattern Recognition Oxford Clarendon Press Box G E P and Cox D R 1964 An analysis of transformations J R Statist Soc B 26 211 252 Chen T and Zhang M Q 1998 POMBE A fission yeast gene finding and exon intron structure prediction system Yeast 14 701 710 Davuluri R Grosse I and Zhang M Q 2001 Computational identification of promoters and first exons in the human genome Nature Genet 29 412 417 Fisher R A 1936 The use of multiple measurements in taxonomic problems Ann Eugen 7 179 188 Fukunaga K 1990 Introduction to Statistical Pattern Recognition 2nd Edition Academic Press San Diego International Human Genome Sequencing Consortium 2001 Initial
311. stry 25 8234 8244 Xu H Wei H Tassone F Graw F Gardiner K and Weissman S 1995a Search for genes from the dark band region of chromosome 21 Genomics 27 1 8 Xu Y Mural R J and Uberbacher E C 1995b Correcting sequencing errors in DNA coding regions using a dynamic programming ap proach CABIOS 11 117 124 Contributed by Edward C Uberbacher Doug Hyatt and Manesh Shah Oak Ridge National Laboratory Oak Ridge Tennessee Finding Genes 4 9 15 Supplement 4 Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences Maja Tarailo Graovac and Nansheng Chen Simon Fraser University Burnaby British Columbia Canada ABSTRACT RepeatMasker is a popular software tool widely used in computational genomics to identify classify and mask repetitive elements including low complexity sequences and interspersed repeats RepeatMasker searches for repetitive sequence by aligning the input genome sequence against a library of known repeats such as Repbase Here we describe two Basic Protocols that provide detailed guidelines on how to use RepeatMasker either via the Web interface or command line Unix Linux system to analyze repetitive elements in genomic sequences Sequence comparisons in RepeatMasker are usually performed by the alignment program cross_match which requires significant processing time for larger sequences An Alternate Protocol describes how to reduce the processing ti
312. stz align file target sequence configuration file For additional options run Nscan_driver p1 with no arguments Current Protocols in Bioinformatics OBTAINING AND INSTALLING N SCAN ON A LOCAL COMPUTER N SCAN 4 0 is open source software Local copies of N SCAN including source code if needed can be obtained under a free license from Washington University through http mblab wustl edu software This protocol describes how to obtain the N SCAN distribution and install it on the most commonly used type of Linux computer N SSCAN can also be compiled from source code for machines based on other architectures or running other versions of the Unix operating system but these advanced procedures are beyond the scope of this unit Necessary Resources Hardware Computer with CPU based on Intel x 86 architecture and running the Linux operating system at least 2 Gb memory 1 GHz processor and 100 Mb of free disk space are recommended Software N SCAN software distribution The distribution can be downloaded from http mblab wustl edu software by clicking the N SCAN Latest Version link 1 Obtain the N SCAN distribution file and place it in the directory where N SCAN should be installed The user must have read write and execute permission in this directory 2 While in the directory containing the distribution enter the following command at the Unix prompt tar xvzf distribution file name If this command is succe
313. sults from the two radio buttons next to return format html or tar file If html is selected the results will be written as an html file If tar file is selected the results will be packed into an archive using the Unix tar protocol For the example here select html to ure Systems 6 Biology RepeatMasker Results RepeatMasker Rejected Your request has been rejected The sequence file is too large for immediate processing You may resubuat 2 in pieces smaller than 100 kb or by using emai return routing from the webpage or go here to obtain a local copy of RepeatMasker Figure 4 10 1 Sequences with length gt 100 kb cannot be processed via the Web interface user is informed by the RepeatMasker to consider alternate methods 4 10 2 Supplement 25 Current Protocols in Bioinformatics SW perc perc perc query position in query matching repeat position in repeat score div del ins sequence begin end left repeat Class family begin end left ID 638 31 6 3 3 1 4 hgi _dna 3 214 22325 C LiMEg LINE Li 5868 216 4 4 359 32 7 13 0 0 8 h hgi8_dna 490 705 21834 MIRb SINE MIR 27 268 0 2 2773 21 0 6 0 1 2 hgi8_ dna 1375 2464 20075 LiMC4a LINE Li 6740 7882 0 3 589 37 1 0 4 1 3 hgl _dna 2598 2832 19707 MIRb SINE MIR 20 252 16 4 493 34 6 3 4 1 6 hgl _ dna 3643 3726 16813 MIR SINE MIR is 97 165 s 378 0 0 0 0 0 0 hgi8_ dna 3727 3768 18771 TA n Simple_repeat 2 4
314. t 25 gt hgl8_dna range chr10 62743355 62765893 S pad 0 3 pad 0 strand repeatMasking none CTNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNACCACTICCTGITGCATTITIGICITICICATITIAA TATGCCAGCTATCTTTICTATTIICCIICICIGGITTATTACCITITATCA TATTIGACTTIGICITICITATTTCAAATCTACTITATIGCAGATGCTAC CTICAGTGTITGATGITATTATITITIATCCITACCCITITAGIGAATICAT TIGCACAGATAAGICTCAAATCCATTICIGIAAGGCCTGICCIGAGIGIG ATTTCTACCTACCTICCTCTCAAAAACAGICGATIGATINNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNN NNNNNCTGAATACCCATTGTAAGITAGGTACAGGGGTAGGTATTAGGAAT TCAAAAATATGGTATCTATCTITTAGGATAATACTICCIGITICICTACTGG AGGTATTTICTATTAACATGICICAATAATICITAAACTAAATATGICAA AACTGAAGTCTATGCTTTICITGACACAGAGICAATCATICCICATATTIC CAGTGGCACCTTATATATTCAGCTCTCTAAGATAACAAACAGAATAATIT TACACTTCCCCCAACCCTCIGICGIGICIGICACTATCICTIAGCCAATTA TITITCICTAATGITITIGCITICICITITITCITICIICIGCIGACACTT TTATTCIGGTAGIGGGCCTTTTTICACTCCATGCATAGGTAGCCTTAACTA GCTATTTTTAGICITCCAGGCTTTTGCCCATTCATCIGITATATCITACG CCACAGCATGAGAATCATCTTGTAACACAATTCCATCACACACACCCCTG CTTAGCTTITATAATATTICTICICTAATACTAGITATACCAGATCCCAACT CCTTAGACTGATGTGCAAAGTACTCTAAATTICCTACCCACTTACTCTCTC CACTCCCATCTCACCAAGGTTAGITCTCATT
315. t dna fa masked current dna fa log Finding Genes 4 10 9 Current Protocols in Bioinformatics Supplement 25 Using RepeatMasker to Identify Repetitive Elements in Genomic Sequences 4 10 10 Supplement 25 current dna fa dna cat current dna fa dna out current dna fa dna tbl The result files are explained in Guidelines for Understanding Results below 8 RepeatMasker provides users with a large array of options to meet the needs ap propriate for different cases Here only commonly used ones are covered For more advanced options users are encouraged to read the help file repeatmasker help which comes with the RepeatMasker program package Note that the order of the command line options is not important when entering multiple commands a Species options and the 1ib flag allow users to specify a particular library file for the corresponding organism RepeatMasker provides common name flags for some species like cat or dog but not for all For that reason usage of Latin names as a species option is highly recommended Users can also provide a repeat library file especially if the library file is not from Repbase collection to RepeatMasker using a lib flag The default repeat library is for primate To establish one s own repeat library for RepeatMasker use the format for IDs as recom mended by repeatmasker help e g gt repeatname tclass subclass or simply gt repeatname tclass b Masking opti
316. t parameters are used Using N SCAN to analyze large scale genomic sequences Annotations of complete genomes N SCAN has been used to annotate all mammalian sequences for which sequence is available at the time of this writing as well as twelve fruit fly species TWIN SCAN was used for gene prediction on the Arabidopsis thaliana maize and rice genomes and on nematode worm species These gene predictions are available through http mblab wustl edu predictions and the UCSC genome browser UNIT 1 4 Details of the way N SCAN is run on mam malian and fly genomes and its performance can be found in Gross and Brent 2006 Ex perimental verification by RT PCR and se quencing has shown that N SCAN can iden tify new genes or additional exons not found in RefSeq Brown et al 2005 Wei and Brent 2006 Additions to N SCAN are a method for adding EST evidence that greatly improves specificity Wei and Brent 2006 and an iter ative masking reprediction method to remove pseudogenes from mammalian gene predic tions van Baren and Brent 2006 The Pub lications page of the Brent laboratory Web site http mblab wustl edu contains refer ences to other papers that have employed TWINSCAN N SCAN in their analyses and whole genome annotations can be downloaded from http mblab wustl edu predictions Current Protocols in Bioinformatics Critical Parameters and Troubleshooting Parameter sets N SCAN relies on a parameter file
317. te score found in the window 54 3 relative to 3 ss using the pre computed weight matrix 4 3 ss splice site score x position dependent triplet preference for true_acceptor versus pseudo_acceptor in the window 24 3 using pre computed 3mer weight matrices 5 Exon_score x average 6mer preference for exon versus intron sum Of Pexon intron over all overlapping 6mers in the exon window exon length 5 6 Strand_score x average Omer exon preference for the forward strand versus the reverse sum of f_exon w f_exon w f_exon w over all overlapping 6mers w in the exon window exon length 5 where the 6mer w is the reverse complement of w 7 Frame_score x max o frame specific 6mer preference for exon versus intron in frame i in the exon window Current Protocols in Bioinformatics Finding Genes 4 2 17 Using MZEF to Find Internal Coding Exons 4 2 18 8 5 ss splice site score xs position dependent triplet preference for true_doner versus pseudo_doner in the window 3 8 using pre computed 3mer weight matrices 9 Exon intron_transition x average exon_preference to the left intron_prefer ence to the right sum Of Pexon intron OVEF all overlapping 6mers in the 54 bp window to the left of 5 ss sum Of Pinton exon OVer all overlapping 6mers in the 54 bp window to the right of 5 ss 49 For the Arabidopsis_MZEF Zhang 19
318. ted by GlimmerM may be an indication of alternative splicing an often overlooked phenomena in the design of gene recognition programs Incorporating other scoring systems While the original GlimmerM system used only IMMs to score potential coding regions the later versions of the system integrate several other scoring methods Based on a method described in Salzberg et al 1998b the authors built a scoring function based on decision trees in order to estimate the probability that a DNA subsequence is coding or not Five types of subsequences are evaluated introns initial ex ons internal exons final exons and single exons Each subsequence is run through ten different decision trees built with the OC1 sys tem Murthy et al 1993 1994 The prob abilities obtained with the decision trees are averaged to produce a smoothed estimate of the probability that the given subsequence is of a certain type In the end as in the malaria ver sion a gene model is accepted only if the IMM score for the coding sequence in the correct reading frame exceeds a fixed threshold Applications of GlimmerM GlimmerM was used as the primary gene finder for chromosome 2 of P falciparum Salzberg et al 1999 and a later version was used to annotate two model plants Arabidopsis thaliana The Arabidopsis Genome Initiative 2000 and the still ongoing project to sequence the rice genome Oryza sativa Yuan et al 2001 It is currently being us
319. ted to text information that contains the coordinates of the different genomic elements Boxes and other elements in the image contain additional information that is shown according to the level of detail selected in the options below the picture Zoom in and out using the 10x 3x and 1 5x buttons Use the scroll buttons to move along the sequence The UCSC genomic features are displayed along the chromosomes Then the geneid GFF output must be adapted to fit the range of the current window The translation of the geneid predictions into the UCSC genomic window can be easily obtained by adding the value 13 903 812 to each coordinate in the geneid output This conversion can be performed with a text editor On Unix systems this operation can be easily performed using any standard file editing tool e g awk or sed For example the awk command would be sawk BEGIN OFS t print 1 2 3 4 13903812 5 413 903812 6 7 8 examplel gt geneid_ucsc gff To add the geneid GFF output press the Add custom tracks button in the UCSC genome browser The geneid output modified in the previous step must be copied into the Data box The following header must be placed just before the geneid GFF predictions Including annotations into the UCSC genome browser browser position chr21 13903812 13935812 track name geneid description geneid predictions visibility 1 color 0 150 50 A new track named geneid will be imported
320. tes in eukaryotic mRNA GlimmerM combines several techniques that have already proven successful in charac terizing the patterns around the donor and ac ceptor sites The splice site predictor algorithm uses a decision tree method called maximal dependence decomposition MDD first intro duced by Burge and Karlin 1997 which is enhanced by Markov models that capture addi tional dependencies among neighboring bases in a region around the splice site This method considers only a small window around the splice junctions which contains most of the information recognized by the spliceosome The authors algorithm also takes advantage of the fact that the coding and non coding se quences switch at the splice junction and this switch can sometimes be detected by consider ing sequence statistics in a larger window In addition by applying the local score optimality feature developed by Brendel and Kleffe 1998 the authors increased the overall per formance of the splice site detection system Using interpolated Markov models to select a gene model Selecting the best gene model depends on a combination of the strength of the splice sites and the score of the exons produced by a special generalization of a Markov chain called an interpolated Markov model IMM Salzberg et al 1999 IMMs are a generalization of fixed order Markov chains The main distinction is that rather than deciding in advance how many bases to consider for ea
321. text editor see APPENDIX 1c Before modifying the values of the thresholds specified on lines 8 9 12 13 and 15 of the config file consult the false positive and false negative rates from the following files false nofilter acc false nofilter don false filter acc false filter don and false atg respec tively These threshold files can be found in the same directory as config file and all of them have the same format Figures 4 4 4 and 4 4 5 present the first lines of the false negative false positive rates for the acceptor and donor sites From Figure 4 4 2 one can see that the default value of the thresholds for the acceptor and donor sites was set to 1 5 41 and 8 76 respectively This corresponds to a 0 false nega tive rate for the acceptor sites and a 0 39 false negative rate for the true donor sites A user might not be satisfied that 6 0 of the GTs in the data will be called donor sites in which case one can set a higher threshold in order to have fewer false predictions For instance a threshold of 1 26 will introduce fewer false positives only 2 0 of all GTs that are not donor sites in the data but 10 or 4 of the true donor sites will be missed This threshold can be introduced in line 9 of the config _ file see Table 4 4 2 to reflect the new rates All threshold parameters from the config_ file lines 8 9 12 13 and 15 see Table 4 4 2 can be changed in the same way by analyzing the corresponding threshold
322. that of GeneMark hmm see Alternate Protocol 1 when using just one model USING GeneMarkS FOR PROKARYOTIC MODEL BUILDING The GeneMarkS program Besemer et al 2001 with the Web interface is shown in Figure 4 5 8 uses a nonsupervised training procedure to build a model for an anonymous DNA sequence It does this through an iterative process beginning with analysis of the sequence using heuristic model building see Basic Protocol 2 Iteratively GeneMarkS predicts coding regions and uses the predicted genes as a training set for the models used in the next iteration The process ends when the predicted coding regions obtained with the updated models do not differ from the previous iteration GeneMarkS also computes an RBS model using the Gibbs Motif Sampler Lawrence et al 1993 to align the upstream DNA sequences of the predicted starts GeneMarkS is efficient for the analysis of large genomes that possess enough coding and noncoding regions to derive accurate higher order models The use of the RBS model leads to a high accuracy of gene start predictions For sequences smaller than 1 Mb the accuracy may deteriorate GeneMarkS can also be used to analyze large phage genomes Most phages use the translational machinery of their hosts Translation signals such as the ribosomal binding site found in prokaryotes will thus be present in the phages that infect those organisms and these signal sequences can consequently be detected using GeneMa
323. the button labeled Browse below the DNA Sequence box not shown in figure allows the user to search the local disk for the desired file to upload 10 Click the Go button below the DNA Sequence box not shown in figure to launch the analysis Alternatively click the Reset form button to reset the form to its default values and repeat steps 1 to 10 Upon launching the analysis the user will be provided with a summary page showing basic information about the request the sequence name the sequence length a request ID and an estimated time to complete the analysis Finding Genes 4 9 3 Current Protocols in Bioinformatics Supplement 4 ALTERNATE PROTOCOL GrailEXP and Genome Analysis Pipeline for Genome Annotation 4 9 4 Supplement 4 11 On the page which then appears click the Check results button to check on the progress of the analysis After clicking on the Check results button the user will be redirected to a self refreshing page which checks every 60 sec for status of the requested tasks Once the job is complete the user will be automatically redirected to the results page Many analyses can be quite time consuming and the user might not want to sit at the workstation and wait for the results to complete GrailEXP therefore assigns each request a Request ID so that the user may retrieve the results at a later date On completion of the analysis results are returned to a Web page w
324. the repeat Position in repeat Begin Starting position of match in repeat consensus sequence End End position of match in repeat consensus sequence Left Number of bases in repeat consensus sequence past the end of the current match ID Repeat identification number Note that if the repeat consensus matches the positive strand the three subcolumns are begin end and left otherwise the three subcolumns are left end and begin The out file Fig 4 10 2 in the Web example is the annotation file that contains the cross_match summary lines The file is basically self explanatory The columns of the out file are described briefly in Table 4 10 1 The matches domains are masked in the masked file This file can be parsed with the help of the BioPerl module Bio Tools RepeatMasker http www bioperl org The masked file Fig 4 10 3 is the same as the query sequence except that the repetitive elements are masked using Ns Xs or lowercase letters if one has a x or xsmall flag on command line or checked the box Mask with Xs or lower case to distinguish masked regions from Ns already in query on the RepeatMasker Web site The tbl file Fig 4 10 4 summarizes the annotation results shown in the out file Notably the tbl file states the percentage repetitive elements coverage COMMENTARY Background Information the Smith Waterman SW alignment algo rithm Smith and Waterman 1981 The pro gram cro
325. this case feature type of exon start and end positions score strand frame and group gene to which the exon belongs 7 Obtain the same output in the XML format using the command geneid P param human3iso param M samples examplel fa Extensible Markup Language or XML http www w3 org XML is a language developed from the experience obtained in the creation of SGML Standard Generalized Markup Language and HTML Hypertext Markup Language which is more widely used on the Internet XML is basically a format to transfer information between computer programs non human readable Many parsing and displaying methods are available for XML which makes it a powerful format to create Web documents geneid supports XML format for predicted genes by means of the option M The DTD Document Type Definition of geneid XML documents can be printed with the option m 8 Examine the complete list of available options by using the option h sgeneid h The most relevant options that have not been discussed are v verbose This produces real time detailed information while geneid is processing the input sequence W C forward reverse This forces prediction in only one strand of the sequence D CDS sequence This prints the DNA coding sequence of predicted genes O R S external features By means of these options additional information can be provided to geneid in order to modify the ab initio prediction see Basic P
326. tion if provided see Basic Protocol 3 Exon mode to predict only signals and exons disabling gene assembling see Basic Protocol 1 steps 3 and 4 or Assembling mode to only assemble the best genes from the external information when provided e g predictions from gene prediction programs other than geneid in GFF format In the DNA Strands menu the user can select where to predict genomic elements For ward and Reverse default Forward positive or Reverse negative 3 Choose the output format and elements to be displayed in the Output Options section Fig 4 3 12 There are two different sets of Output Options those concerning the format and those concerning the elements to display The available formats are GFF geneid extended format and XML as well as a format containing the CDS sequence for each predicted gene for further details about the formats see Basic Protocol steps 2 5 6 and 7 The signals that can be included in the output are Acceptor and Donor splice sites and Start and Stop codons There are five types of exons First Internal Terminal Single and ORFs There is also an option to build an ordered output containing all of the predicted exons see Basic Protocol 1 step 2 for details about the type of genomic elements predicted by geneid 4 Examine the geneid output Fig 4 3 13 The results for the sequence examplel1 fa see Basic Protocol 1 steps 1 and 2 for detailed
327. tions of gene finding programs e g GENSCAN MZEF unir 4 2 to create more reliable annotations is also presented see Support Protocol USING WEB BASED FirstEF TO PREDICT PROMOTERS AND FIRST EXONS The user can submit a FASTA formatted DNA sequence APPENDIX 1B to FirstEF either through the World Wide Web hittp rulai cshl org tools FirstEF as described in this proto col or through a locally installed version see Alternate Protocol Since the Web server has a restriction on the size of the input file with the maximum being 100 kb it is advisable to obtain a local copy of the FirstEF software for analyzing large genomic sequences e g the entire DNA sequence of a human chromosome Necessary Resources Hardware Computer with Internet access e g PC running Microsoft Windows or Linux Apple Macintosh Unix workstation Software Internet browser e g Netscape Navigator Microsoft Internet Explorer Files DNA sequences to be analyzed in FASTA format APPENDIX 1B Sample sequences can be found at the FirstEF Web site The example Example 1 used in this unit isa DNA sequence of length 100 kb from human chromosome 20 chr20 300001 400000 NCBI build 30 which can be found at the Current Protocols Web site hittp www3 interscience wiley com c_p cpbi_sampledatafiles htm Submit sequence to the FirstEF Web server 1 Access the FirstEF Web server through an Internet browser http rulai cshl org tools FirstEF Upload t
328. tocol Another advantage of having GlimmerM locally installed is that the parameters of the system can be customized to reflect the user s expertise about the organism e g by changing default parameters of the program such as the minimum gene length or the prediction overlap allowed Necessary Resources Hardware A Unix workstation GlimmerM has been successfully compiled for Linux Digital Unix and SunOS and it should be easy to compile on any platform supporting ANSI C and C Finding Genes Contributed by Mihaela Pertea and Steven L Salzberg 4 4 1 Current Protocols in Bioinformatics 2003 4 4 1 4 4 20 Copyright 2003 by John Wiley amp Sons Inc Using GlimmerM to Find Genes in Eukaryotic Genomes 4 4 2 Software Currently there are two packages available GlimmerM 1 2 and GlimmerM 2 0 In the GlimmerM 1 2 package the code of the gene finder is trained and customized specifically for each organism GlimmerM 2 0 is upgraded to contain the automatic training procedure see Support Protocol and a generally applicable gene finding algorithm GlimmerM 2 0 contains all of the organism specific versions found in version 1 2 however the performance of these versions is slightly different due to changes in parameter settings when building the later system The example below uses GlimmerM 2 0 but the basic procedure for running versions 1 2 and 2 0 is the same A truly determined user who is studying all of
329. tory regions CpGpromoter oshikhes and Zhang 2000 is another pro moter prediction program for a large scale hu man promoter mapping using CpG islands Literature Cited Burge C and Karlin S 1997 Prediction of com plete gene structures in human genomic DNA J Mol Biol 268 78 94 Davuluri R V Grosse I and Zhang M Q 2001 Computational identification of first exons and promoters in human genome Nature Gen 29 412 417 Finding Genes 4 7 9 Supplement 1 Application of FirstEF to Find Promoters and First Exons in the Human Genome 4 7 10 Supplement 1 Florea L Hartzell G Zhang Z Rubin G M and Miller W 1998 A computer program for align ing a cDNA sequence with a genomic DNA sequence Genome Res 8 967 974 Gardiner Garden M and Frommer M 1987 CpG islands in vertebrate genomes J Mol Biol 196 261 276 Ioshikhes I and Zhang M Q 2000 Large scale human promoter mapping using CpG islands Nature Gen 26 61 63 Scherf M Klingenhoff A and Werner T 2000 Highly specific localization of promoter regions in large genomic sequences by PromoterInspec tor A novel context analysis approach J Mol Biol 297 599 606 Zhang M Q 1997 Identification of protein coding regions in the human genome by quadratic dis criminant analysis Proc Natl Acad Sci U S A 94 565 568 Key References Davuluri et al 2001 See above The algorithm details of FirstEF
330. ttempt to model spliced messenger RNAs Other attempts to identify coding regions have used dynamic pro gramming and neural networks Geneparser Snyder and Stormo 1993 oligonucleotide composition and discriminant analysis Solovyev et al 1994 and linguistic methods Genlang Dong and Searles 1994 DNA dialects vary among different organ isms Parameters such as the frequency with Current Protocols in Bioinformatics which individual codons are used the length of introns and exons and the amount of repetitive DNA differ for different organisms The GrailEXP system has been optimized for hu man DNA though it appears to function rea sonably well for most mammalian species As more distantly related organisms are studied it is necessary to construct specialized systems for each organism GrailEXP now supports the analysis of sequences from Drosophila melanogaster Arabidopsis thaliana and the mouse If the coding exon recognition portion of the GRAIL 2 system performed perfectly as sembling a model of the spliced mRNA i e the gene model would be trivial involving nothing more than connecting the ends of the predicted exons However because the coding region prediction is less than perfect a compu tational method must be used to test various combinations of exons in order to propose a gene model The gene assembly program Gawain in GrailEXP uses dynamic program ming to assemble gene models from the pre dict
331. ty to each other The first exon of gene 1 will often be joined with the second exon of gene 2 and so on A future improvement being considered is a module for recognition of repeat gene regions which would identify and suit ably handle such occurrences Galahad can identify exons as short as 10 bases by identifying suitable splice sites to which to align while processing a much longer BLAST alignment GrailEXP finds short exons Current Protocols in Bioinformatics as short as 5 bases at the edges of the align ments provided the bases in the message ex actly match the genomic sequence This is criti cal for aligning the CDS portions of transcripts with genomic sequence where there could be very short exons at the edges of the gene A future improvement being considered is robust recognition of short exons in which the possible presence of such exons is first determined by detecting a bump in the alignment and a high donor splice site score The current implemen tation does place the alignment pieces in proper frames so missing a short exon will only cause a slight ripple in the resulting translation It is possible to use exon predictions from another exon or gene finding system like GENESCAN as input to Galahad In fact Galahad can interpret GENESCAN output files Additionally exons from any other pro gram can be used by formatting the exon in formation into the GrailEXP exon format or GENESCAN output format Ad
332. uences Detection of genes in sequences from prokaryotic organ isms by GeneMark hmm is described in unr 4 5 The eukaryotic GeneMark hmm uses Markov models of protein coding and noncoding sequences as well as positional nucleotide frequency matrices for prediction of the translational start translational termination and splice sites All these models along with length distributions of exons introns and intergenic regions are integrated into one hidden Markov model The algorithm implemented in GeneMark hmm finds the maximum likelihood path of this model through hidden coding and noncoding states given the analyzed sequence The GeneMark program Borodovsky and MclIninch 1993 also see UNIT 4 5 may be run in conjunction with GeneMark hmm to provide additional insight into how the DNA sequence is structured in terms of coding potential These GeneMark hmm and GeneMark programs are accessible via the Internet see Basic Protocol at http opal biology gatech edu GeneMark Alternatively local versions of the software are available which can be run under the Unix operating system see Alternate Protocol 1 The programs on the Web site are revised as soon as updates are available therefore the users of the Web site normally have access to the latest version of the software A modified version of GeneMarkS Besemer et al 2001 can be used to detect genes in eukaryotic viruses see Alternative Protocol 2 This program utilizes a nonsupervise
333. ults of GlimmerM are printed on the screen but the user can redirect the output of the system by using a gt sign followed by a filename Current Protocols in Bioinformatics Finding Genes 4 4 3 Table 4 4 1 Optional Parameters to Use When Running GlimmerM Command Argument Default values Version 1 2 Version 2 0 d dir J J Set the directory of the training files to dir g n 175 60 Set minimum gene length to n o n 30 0 Set minimum overlap length to n nucleotides Overlaps shorter than this are ignored p n 10 0 Set minimium overlap percentage to n Overlaps shorter than this percentage of both strings are ignored n 99 90 Set the threshold score above which a DNA region is called a gene to n If the in frame score is greater or equal to n an integer between 0 and 100 then the region is given a number and considered a potential gene r Do not use independent probability score column No No r Use independent probability score column Yes Yes f Do not use maximal local filtering of the splice sites n a No f Use maximal local filtering of the splice sites n a Yes S Do not use the translation start site model n a No S Use the start site model Yes Yes 5 t n a Setinconfig file Use threshold t for the acceptor sites 3 t n a Setinconfig file Use threshold ft for the donor sites Further discussion of these parameters can be found in the Critical Parameters section gt The o and p o
334. und in the query DNA sequence See Guidelines for Understanding Results for explanation 6 Select one of the entries from the pull down menu next to DNA source each of which corresponds to a different repetitive element library The default is Human For the example here select Human because the sequence is from the human genome Note that if the query sequence is from an organism that is not listed here the command line version of RepeatMasker must be run locally see Basic Protocol 2 and an appropriate repeat file from Repbase Update must be used if there is one If working with a genome for which Repbase does not have an appropriate repeat library RECON Bao and Eddy 2002 Stein et al 2003 or RepeatScout http bix ucsd edu repeatscout Price et al 2005 can be used to establish one from scratch 7 In the series of pull down menus radio buttons and check boxes under Lineage Annotation Options select the appropriate options These options are self explanatory For example if Comparison Species is selected the lineage specific repeats are annotated with the RepeatMasker output with respect to the selected species Current Protocols in Bioinformatics Finding Genes 4 10 5 Supplement 25 c c c c c Matrix Unknown Transitions transversions 3 27 49 15 Gap_init rate 0 03 7 211 avg gap size 1 43 10 7 636 31 57 3 30 1 39 hgi _dna 3 214 20930 C LiM
335. veloped for parallel model learning and genomic sequence annotation The models learned in this process could be diversified to accommodate the sets of so called Typical Highly Typical and Atypical genes that can be selected in a given bacterial genome Hayes and Borodovsky 1998a For instance in the case of the E coli genome the Highly Typical and Atypical genes correspond to Highly Expressed and Horizontally Trans ferred genes respectively The accuracy of gene prediction can be improved even further if an accurate model of the RBS signal is developed and taken into account Hayes and Borodovsky 1998b GeneMark hmm The GeneMark hmm algorithm Lukashin and Borodovsky 1998 was designed to im prove gene prediction quality in terms of find ing exact gene boundaries The previously de veloped GeneMark program identified a gene mainly as the open reading frame where the gene is residing However the 5 boundary of the gene the translation initiation codon asso ciated with the protein amino terminus might not be precisely predicted The range of uncer tainty for the initiation codon position is of the size of GeneMark sliding window i e 100 nt In fact GeneMark indicates several possible start codons and scores them The underlying idea of GeneMark hmm was to embed the GeneMark models for coding and noncoding regions into the naturally derived hidden Markov model HMM framework with gene boundaries modeled as tra
336. verage sensitivity and specificity above 90 Hence the predictions with CpG window and probability values higher than 0 9 are very likely to be real In case of non CpG related first exons the accuracy of FirstEF is relatively low with average sensitivity and specificity just above 70 In other words FirstEF may miss roughly 3 out of 10 real non CpG related first exons and 3 out of 10 non CpG related first exons may be false predictions that are actually not real First Exon Clusters If two first exon predictions are separated by lt 1000 bp FirstEF considers them as probable first exons of the same gene and places them in the same cluster If a cluster has more than three predictions at least one of those in that cluster is highly likely to be a real first exon Other predictions in the cluster may be alternative first exons which may need additional support such as a cDNA EST match Predictions on the Alternative Strand FirstEF predicts first exons on both positive and negative strands If there is a strong promoter region particularly in CpG islands FirstEF tends to predict an overlapping first exon on the opposite strand due to a strong donor site In such cases the predictions that have downstream cDNA EST match should be accepted If there is no supporting cDNA EST match on both sides of the predicted cluster then the one that has the higher probability values should be considered The predictions of FirstEF on the alternative st
337. vsky M Rudd K E and Koonin E V 1996 Metabolism and evolution of H influenzae deduced from whole genome com parison to E coli Curr Biol 6 279 291 Tomb J White O Kerlavage A R Clayton R A Sutton G G Fleischmann R D Ketchum K A Klenk H P Gill S Dougherty B A Nelson K Quackenbush J Zhou L Kirkness E F Peterson S Loftus B Richarson D Dodson R Khalak H G Glodek A McKen ney K Fitzegerald L M Lee N Adams M D Hickey E K Berg D E Gocayne J D Utterback T R Peterson J D Kelley J M Cotton M D Weidman J M Fujii C Bow man C Watthey L Wallin E Hayes W S Borodovsky M Karp P D Smith H O Fraser C M and Venter J C 1997 The complete genome sequence of the gastric pathogen Heli cobacter pylori Nature 388 539 547 Internet Resources http opal biology gatech edu GeneMark GenMark Web site Contributed by Mark Borodovsky School of Biology and School of Biomedical Engineering Georgia Institute of Technology Atlanta Georgia Ryan Mills John Besemer and Alex Lomsadze School of Biology Georgia Institute of Technology Atlanta Georgia Current Protocols in Bioinformatics Eukaryotic Gene Prediction Using GeneMark hmm In this unit eukaryotic GeneMark hmm Lukashin and Borodovsky 1998 M Boro dovsky and A V Lukashin unpub observ is presented as a method for detecting genes in eukaryotic DNA seq
338. wise it proceeds to next GT For every candidate donor site FirstEF scans a region of length 2000 nt 1500 nt upstream and 500 nt downstream of GT for the existence of a CpG window of length 201 nt with a CpG score 6 5 FirstEF decides whether the first exon is CpG related or non CpG related de pending on the presence or absence of a CpG window For more discussion of CpG window and CpG score see Davuluri et al 2001 FirstEF uses a sliding window of length 570 nt considering the first 500 nt as a proximal promoter upstream of transcription start site TSS and the following 70 nt as downstream of TSS within the 1500 nt upstream region of each candidate donor site FirstEF decides whether the sliding window can be a promoter or not based on the a posteriori probability of promoter P promoter P promoter was calcu lated using two different promoter QDFs Pro moter QDF one for CpG related and the other for non CpG related If P promoter promoter cut off value FirstEF matches the promoter region with the corresponding donor site and evaluates the a posteriori probability of exon P exon by us Current Protocols in Bioinformatics ing four different first exon QDFs exon QDF FirstEF reports all those exons with P exon exon cut off value along with the promoter region and CpG window if it exists The user can select different cut off values for donor promoter and exon in the range of 0 2 to 1 Critical Par
339. wo vectors x y within class k the discriminant function will be a quadratic function of x through A defined in Equation 4 2 2 1 De E ara AP yl len 4 2 3 where y In n n_ Geometrically the decision boundary is a quadratic hyper surface in p dimensions Figure 4 2 7 when amp X_ Using such a quadratic discriminant function for classification is called QDA quadratic discriminant analysis When X_ X the quadratic terms in h x will be canceled out Current Protocols in Bioinformatics Finding Genes 4 2 15 Xy A Figure 4 2 7 Quadratic decision boundary for normal distributions Xk Figure 4 2 8 Linear decision boundary for normal distributions when 2 X_ h x u 4 Stes En u E u_ y 4 2 4 The Bayes decision boundary will become linear hyper plane as seen in Figure 4 2 8 Although linear decision boundaries are optimal in the Bayes sense only for normal distributions with equal covariance matrices because of its simplicity one may always want to know how well one can do with just a linear discriminant function for an arbitrary class of distributions A general linear discriminant function can be written as h x V x v which means x is projected onto a vector V and the variable y V x in the projected linear space is classified according to whether y gt v or y lt v Suppose the means and variances in the projected subspace are
340. words need to be bracketed by quotation marks e g Caenorhabditis elegans Other than the w option which indicates that WU BLAST is used the command line parameters and options are similar to those in Basic Protocol 2 GUIDELINES FOR UNDERSTANDING RESULTS The output of RepeatMasker is written into five different files in the same directory where the query sequence or sequences reside Only three files those with out masked and tbl extensions contain results others store processing information and are therefore not detailed here If RepeatMasker is run via the Web server interface the contents of these three files are written into one page file shown in Figures 4 10 2 4 10 3 and 4 10 4 respectively Current Protocols in Bioinformatics Table 4 10 1 Columns of the out File from Left to Right also see Fig 4 10 2 Column Content SW score Smith Waterman score of the match Perc div Percent substitutions in matching region compared to the consensus Perc del Percent of bases opposite a gap in the query sequence deleted bp Perc ins Percent of bases opposite a gap in the repeat consensus inserted bp Query sequence Position in query Name of query sequence Begin Starting position of match in query sequence End End position of match in query sequence Left Number of bases in query sequence past the end position of the current match Matching repeat Repeat Name of repeat Class family The class of
341. wx mzef_cmd To get a description of the parameter entry order type in the command name by itself and MZEF will output a short usage snippet smzef cmd Usage mzef_cmd seqfile strand p0 overlap sequence file in fasta format required strand 1 default forward 2 reverse p0 prior probability default 0 04 overlap maximum exon overlap default 0 See Critical Parameters for further discussion of these parameters Run the command line version on the local computer smzef cmd m12523 fasta 1 0 02 1 The results will be printed to the screen Here Overlap 1 is entered and therefore one can see there are several overlapping exons in the output see Figure 4 2 2 USING THE INTERACTIVE UNIX VERSION MZEF TO ANALYZE GENOMIC DNA SEQUENCES Necessary Resources Hardware Any Unix or Linux workstation Software The appropriate MZEF interactive executable file e g mzef The executable files for MZEF are free for academic users The files may be downloaded from the cshl org FTP site see step 1 below Commercial users and those who wish to obtain source codes written in FORTRAN 77 should contact the CSHL licensing office Dr Carol Dempster 516 367 6885 dempster cshl org The software has evolved into many different versions to meet the demands from different users Consequently there are several executable files available from the FTP site The file names indicate the differences between the various forms
342. xons in the DNA sequence are known the user can pass them to geneid via the R option see Basic Protocol 3 If the user suspects that whole exons or genes have been missed one can modify some of the values of the parameter file to attempt to recover them There are two reasons why ex ons or genes may have been completely missed by geneid Either 1 geneid does not consider them as candidate exons or 2 it does predict them as candidate exons but they are not in cluded in the final gene prediction It is easy to check which is the case by using the x option which outputs the complete list of can didate exons predicted by geneid In the sec ond case the user can increase the value of the Exon Weight EW parameter in the parameter file see Background Information By default this number is negative for most species The higher the value the higher the number of ex ons included in the final gene prediction If the missing exons have not been included in the list of candidate exons then decrease the cutoff values of exons and sites and probably still increase the value of EW Finding Genes 4 3 25 Supplement 19 Using geneid to Identify Genes 4 3 26 Supplement 19 If there is biochemical or other evidence suggesting that the sequence encodes only a single gene the authors suggest that you use a gene model that also reflects a single exon gene geneid runs correctly but stops with a warning before pro
Download Pdf Manuals
Related Search
Related Contents
MANUAL DE INSTRUÇÕES FUTABA HI-4460 Graphical Readout User Manual - ETS Kenmore 7.0 cu. ft. Electric Dryer w/ Wrinkle Guard - White Owner's Manual Toshiba Portégé R930-S9331 DESIGN PACKAGE Copyright © All rights reserved.
Failed to retrieve file