RaxML user manual - Trex
Contents
1. u multiBootstrapSearches SaaS SSS a SSS a lt q wo w workingDirectory x rapidBootstrapRandomNumberSeed y z multipleTreesFile N number0fRuns Depending on the compiler you used and the platforms that are at your disposal you will have three alter native executables 1 raxm1HPC is just the sequential version 2 raxmlHPC MPI is the parallel coarse grained version It can be used if you have a LINUX cluster available and want to perform multiple analysis or multiple rapid bootstraps i e in combination with the N or N and b x or f a x options Note that if you do not specify N 3 raxm1HPC Pthreads only makes sense if you have access to a shared memory or multi core machine Note that N can be used as an alternative to since the character seems to cause problems with some parallel job submission systems because it is sometimes used to start comments The options in brackets are optional i e must not be specified whereas RAxML must be provided the sequence file name with s and the output file s name appendix with n and the desired model of DNA or AA substitution with m Let s have a look at the individual options now a weightFileName This option specifies the name of a column weight file which allows you to assign individual weights to each column of the alignment The default is that each column has the weight 1 The weights in the weight file must be integers2. know e ML model for morphological binary data e ML based rapid sequence addition option e More efficient ML function implementation for very gappy multi gene alignments For any further requests or suggestions please send an email to stamatakis bio ifi 1mu de or con tact me via skype internet telephony login stamatak Acknowledgments Many people have contributed to improve RAxML either via personal discussions email or skype or by pro viding real world alignments and answering all sorts of CS and biology related questions In the hope not to have forgotten anybody would like to thank the following colleagues names are in no particular order Ziheng Yang Olivier Gascuel Stephane Guindon Wim Hordijk Michael Ott Olaf Bininda Emonds Maria Charalambous Pedro Trancoso Tobias Klug Derrick Zwickl Jarno Tuimila Charles Robertson Daniele Catanzaro Daniel Dalevi Mark Miller Usman Roshan Zhihua Du Markus Goker Bret Larget Josh Wilcox Marty J Wolf Aggelos Bilas Alkiviadis Simeonidis Martin Reczko Gangolf Jobb Frank Kauff James Munro Peter Cordes Tandy Warnow Bernard Moret Paul Hoover Jacques Rougemont Joe Felsenstein Daniel Lundin References 1 Felsenstein J Evolutionary trees from DNA sequences a maximum likelihood approach Journal of Molecular Evolution 17 1981 368 376 2 Felsenstein J Phylip phylogeny inference package version 3 6 2004 Distributed by the author Department of Genome Scie
3. L 0 95 n TEST m model0fEvolution Selection of the model of nucleotide substitution or amino acid substitution to be used NUCLEOTIDE MODELS m GTRCAT GTR approximation with optimization of individual per site substitution rates and classifi cation of those individual rates into the number of rate categories specified by c This is only a work around for GTRGAMMA so make sure not to compare alternative topologies based on their GTRCAT likelihood values Therefore you can not use GTRCAT in combination with f e tree evaluation and not in combination with multiple analyses on the original alignment N option This is due to the fact that the author assumes that you want to compare trees based on likelihoods if you do a multiple run on the original alignment If you specify e g m GTRCAT and 10 the program will automatically use GTRMIX see below 10 m GTRMIX This option will make RAxML perform a tree inference search for a good topology under GTRCAT When the analysis is finished RAxML will switch its model to GTRGAMMA and evaluate the final tree topology under GTRGAMMA such that it yields stable likelinood values m GTRGAMMA GTR General Time Reversible model of nucleotide substitution 17 with the r model of rate heterogeneity 19 All model parameters are estimated by RAxML The GTRGAMMA implemen tation uses 4 discrete rate categories which represents an acceptable trade off between speed and accuracy Note that
4. site log Likelihoods for one ore more trees passed via z They will be written to a Treepuzzle formatted file 21 that can be read by CONSEL 22 Example raxmlHPC f g s alg m GTRGAMMA z trees n TEST f h RAxML will compute a log likelihood test SH test 23 between a best tree passed via t and a bunch of other trees passed via z Example raxmlHPC f h t ref z trees s alg m GTRGAMMA n TEST f i performs a really thorough standard bootstrap in combination with b option DOES NOT WORK with x RAxML will refine the final BS tree under GAMMA and a more exhaustive algorithm Example raxmlHPC f i b 12345 100 s alg m GTRCAT n TEST f j generates a bunch of bootstrapped alignment files from an original alignment file Example raxmlHPC f j b 12345 100 s alg m GTRCAT n TEST fm RAxML will compare bipartitions between two bunches of trees passed via t and z respectively The program will return the Pearson correlation between all bipartitions found in the two tree files A file called RAxML_bipartitionFrequencies outpuFileName will be printed that contains the pair wise bipartition frequencies of the two sets Example raxmlHPC f m t treesi z trees2 s alg m GTRCAT n TEST f n computes the log likelihood score of all trees contained in a tree file provided by z under GAMMA or GAMMA P Invar Example raxmlHPC f n z trees s alg m GTRGAMMA n TEST f o RAxML will execute the slower old search
5. 5 2 3 Finding the Best Known Likelihood tree BKL As already mentioned RAxML uses randomized MP starting trees in which it initiates an ML based opti mization Those trees are obtained by using a randomized stepwise addition sequence to insert one taxon after the other into the tree When all sequences have been inserted a couple of subtree rearrangements also called subtree pruning re grafting with a fixed rearrangement distance of 20 are executed to further improve the MP score The concept to use randomized MP starting trees in contrast to the NJ Neighbor Joining starting trees many other ML programs use is regarded as an advantage of RAxML This allows the program to start ML optimizations of the topology from a distinct starting point in the immense topological search space each time Therefore RAxML is more likely to find good ML trees if executed several times This also allows you to build a consensus tree out of the final tree topologies obtained from each indi vidual run on the original alignment By this and by comparing the final likelihoods you can get a feeling on how stable prone to get caught in local maxima the search algorithm is on the original alignment Thus if you have sufficient computing resources available in addition to bootstrapping you should do multiple inferences I executed 200 inferences in some recent real world analyses with Biologists with RAxML on the original alignment On smaller datasets it will also be
6. RAxML_bootstrap MultipleBootstrap 9 you can easily concatenate them by using the LINUX UNIX eat command e g cat RAxML_bootstrap MultipleBootstrap gt RAxML_bootstrap All In order to get a tree with bootstrap values on it just execute RAxML as indicated below raxmlHPC f b m GTRCAT s ex_al z RAxML_bootstrap All t RAxML_result MultipleOriginal RUN 99 n BS_TREE The new output tree format now shows the support values as inner node labels and also displays branch lengths it can look e g like this Human 0 555 Frog 0 207 Carp 0 129 Loach 0 192 100 0 159 70 0 001 Chicken 0 561 100 0 259 65 0 091 Whale 0 108 Cow 0 116 Seal 0 186 55 0 030 65 0 046 95 0 144 Rat 0 068 0 045 Mouse 0 045 19 6 Frequently Asked Questions Q Can use NEXUS style input files for analyses with RAxML Not directly but my colleague Frank Kauff fkauff rhrk uni k1l de at the University of Kaiserslautern has written a cool biopython wrapper called PYRAXML2 This is a script that reads nexus data files and prepares the necessary input files and command line options for RAxML You can download the Beta version of PYRAXML2 at http www lutzonilab net downloads Q Why don t you like the proportion of Invariable P Invar Sites estimate despite the fact that you implemented it only implemented P Invar in RAXML to make some users happy but still strongly disagree with its usage It is unquestionable that one needs to incor
7. about the model and algorithm used and how RAxML was called The final GTRGAMMA likelinood s only if m GTRGAMMA or m GTRMIX have been used as well as the alpha shape parameter s are printed to this file In addition if the rearrangement setting was determined automatically i has not been used the rearrangement setting found by the program will be indicated RAXxML_parsimonyTree exampleRun contains the randomized parsimony starting tree if the program has not been provided a starting tree by t However this file will not be written if a multiple bootstrap is executed using the and b options RAxML_randomTree exampleRun contains the completely random starting tree if the program was exe cuted with d RAxML_checkpoint exampleRun checkpointNumber Printed if you specified by j that checkpoints shall be written Checkpoints are numbered from 0 to n where n is the number of iterations of the search algorithm Moreover the checkpoint files are additionally numbered if a multiple inference on the original alignment has been specified using Writing of checkpoint files is disabled when a multiple bootstrap is executed RAXxML_bootstrap exampleRun If a multiple bootstrap is executed by and b or x all final boot strapped trees will be written to this one single file RAXML_bipartitions exampleRun If you used the f b option this file will contain the input tree with confidence values from 0 to 100 drawn on it It is also pr
8. can be computed from the non diagonal entries You still have to specify an AA substitution model via m to tell the program that it has to read and analyze an AA alignment It will just extract this information from the respective string however by specifying e g m PROTGAMMAWAGF it will use empirical base frequencies instead of the frequencies specified in file proteinModel Example raxmlHPC s alg m PROTGAMMAWAG p proteinModel n TEST q multipleModelFileName This allows you to specify the regions of your alignment for which an individual model of nucleotide substi tution should be estimated This will typically be useful to infer trees for long in terms of base pairs multi gene alignments If e g m GTRGAMMA is used individual a shape parameters GTR rates and empirical base frequencies will be estimated and optimized for each partition 12 If you have a pure DNA alignment with 1 000bp from two genes gene1 positions 1 500 and gene2 positions 501 1 000 the information in the multiple model file should look as follows DNA genel 1 500 DNA gene2 501 1000 If gene1 is scattered through the alignment e g positions 1 200 and 800 1 000 you specify this with DNA genel 1 200 800 1 000 DNA gene2 201 799 You can also assign distinct models to the codon positions i e if you want a distinct model to be esti mated for each codon position in gene1 you can specify DNA geneicodoni 1 500 3 DNA gen
9. implementation described in 13 Finally if you used the CAT approximation of rate heterogeneity see Section 2 2 in your analyses please also cite Alexandros Stamatakis Phylogenetic Models of Rate Heterogeneity A High Performance Computing Perspective in Proceedings of IPDPS2006 14 In case that you use RAxML as a component of larger software packages or Bioinformatics pipelines would greatly appreciate if you could add a text box or analogous appropriate information that RAxML should also be cited separately when used 2 IMPORTANT WARNINGS 2 1 RAXML Likelihood Values It is very important to note that the likelihood values produced by RAxML can not be directly compared to likelihood values of other ML programs However the likelihood values of the current version are much more similar to those obtained by other programs with respect to previous releases of RAXML usually be tween 1 0 log likelihood units of those obtained e g by PHYML IQPNNI 15 or GARLI Note that the deviations between PHYML RAXML and GARLI likelihood values can sometimes be larger because GARLI uses a Slightly different procedure to compute empirical base frequencies Derrick Zwickl personal com munication while the method in RAxML is exactly the same as implemented in PHYML These deviations between RAxML PHYML on the one side and GARLI on the other side appear to be larger on long multi gene alignments Also note that likelihood values obta
10. option will just make RAxML write a reduced alignment file without the excluded columns that can then be used for the real analysis If you use a mixed model an appropriately adapted model file will also be written Example raxmlHPC E excludeFile s alg m GTRCAT q part n TEST In this case the files with columns excluded will be named alg excludeFile and part excludeFile f algorithm This option allows you to select the type of algorithm function you want RAxML to execute f a tell RAxML to conduct a rapid Bootstrap analysis and search for the best scoring ML tree in one single program run Example raxmlHPC f a s alg x 12345 100 m GTRCAT n TEST f b when this is specified RAxML will draw the bipartitions using a bunch of topologies typically boot strapped trees specified with z see below onto a single tree topology specified by t typically the best scoring ML tree Example raxmlHPC f b t ref z trees m GIRCAT s alg n TEST f c just checks if RAxML can read the alignment Example raxmlHPC f c t m GTRCAT s alg n TEST f d DEFAULT RAxML will execute the new as of version 2 2 1 and significantly faster rapid hill climbing algorithm 5 f e RAxML will optimize the model parameters and branch lengths of a topology provided via the t option under GTRGAMMA or the respective AA substitution model under GAMMA Example raxmlHPC f e t ref m GTRGAMMA s alg n TEST f g used to compute the per
11. original pruning point If you don t specify i a good initial rearrangement setting will automatically be determined by RAxML see Section 5 2 1 for further details j Specifies that RAxML shall write intermediate trees found during the search to a separate file after each iteration of the search algorithm The default setting i e if you do not specify j is that no checkpoints will be written k Specifies that RAxML shall optimize branches and model parameters on bootstrapped trees as well as print out the optimized likelihood Note that this option only makes sense when used with the GTRMIX or GTRGAMMA models or the respective AA models q Specify a threshold for sequence similarity clustering RAxML will then print out an alignment to a file called sequenceFileName reducedBy threshold that only contains representative sequences for the in ferred clusters The specified threshold must be between 0 0 and 1 0 RAxML uses the QT clustering algorithm 24 to perform this task In addition a file called RAxML_reducedList outputFileName will be written that contains clustering information This option is turned off by default Example raxmlHPC s alg m GTRCAT 1 0 95 n TEST L Same functionality as 1 above but uses a less exhaustive and thus faster clustering algorithm This is intended for very large datasets with more than 20 000 30 000 sequences and also turned off by default Example raxmlHPC s alg m GTRCAT
12. s ex_al t RAxML_parsimonyTree ST4 n FI4 and then using the automatically determined setting on the same starting trees raxmlHPC f d m GTRMIX s ex_al t RAxML_parsimonyTree STO n AIO raxmlHPC f d m GTRMIX s ex_al t RAxML_parsimonyTree ST4 n AI4 Here we use the GTRMIX model i e inference under GTRCAT and evaluation of the final tree under GTRGAMMA such that we can compare the final likelihoods for the fixed setting FIO FI4 and the automatically determined setting AIO AT4 The setting that yields the best likelihood scores should be used in the further analyses 5 2 2 Getting the Number of Categories right Another issue is to get the number of rate categories right Due to the reduced memory footprint and significantly reduced inference times the recommended model to use with RAxML on large dataset is GTRMIX if you are doing runs to find the best known ML tree on the original alignment and GTRCAT for bootstrapping Thus you should experiment with a couple of c settings and then look which gives you the best T likelihood value Suppose that in the previous Section 5 2 1 you found that automatically determining the rearrangement setting works best for your alignment You should then re run the analyses with distinct c settings by increments of e g 15 rate categories e g raxmlHPC f d c 10 m GTRMIX s ex_al t RAxML_parsimonyTree STO n C10_0 raxmlHPC f d c 10 m GTRMIX s ex_al t RAxML_parsimonyTree ST4 n C10_4
13. separated by any type and number of whitespaces within a separate file In addition there must of course be as many weights as there are columns in your alignment The contents of an example weight file would look like this BLL2ZLI1TLTALi 21tiasgttirtiatidti 1 tTititaitirttrtrttititit4iitir 4it Example raxmlHPC a wgtFile s alg m GTRCAT n TEST b bootstrapRandomNumberSeed This option allows you to turn on non parametric bootstrapping 7 To allow for reproducibility of runs in the sequential program you have to specify a random number seed e g b 123476 Note however that parallel bootstraps with the parallel version raxm1HPC MPI are not reproducible despite the fact that you specify a random number seed They are also not reproducible for the sequential version in case you do not provide a fixed starting tree with t or a parsimony random seed via p Example raxmlHPC b 12345 100 s alg m GTRCAT n TEST c numberOfCategories This option allows you to specify the number of distinct rate categories into which the individually optimized rates for each individual site are thrown under m GTRCAT The results in 14 indicate that the default of c 25 works fine in most practical cases Example raxmlHPC c 40 s alg m GTRCAT n TEST d This option allows you to start the RAxML search with a complete random starting tree instead of the default Maximum Parsimony starting tree On smaller datasets around 100 200 taxa
14. this has been hard coded for performance reasons i e the number of discrete rate categories can not be changed by the user m GTRCAT_GAMMA Inference of the tree with site specific evolutionary rates However here rates are categorized using the 4 discrete GAMMA rates following a formula proposed by Yang 19 Evaluation of the final tree topology is done under GTRGAMMA This option is more for experimental purposes than for everyday use m GTRGAMMAI Same as GTRGAMMA but with estimate of proportion of invariable sites 25 though still don t like the idea see discussion in Section 6 m GTRMIXI Same as GTRMIX but with estimate of proportion of invariable sites m GTRCAT_GAMMAI Same as GTRCAT_GAMMA but with estimate of proportion of invariable sites AMINO ACID MODELS Available AA models Values for matrixName see below DAYHOFF 26 DCMUT 27 JTT 28 MTREV 29 WAG 80 RTREV 31 CPREV 82 VT 33 BLOSUM62 34 MTMAM 35 With the optional F appendix you can specify if you want to use empirical base frequencies Please note that for mixed models you must in addition specify the per gene AA model in the mixed model file see q option below m PROTCATmaitrixName F AA matrix specified by matrixName see above for a list with optimization of individual per site substitution rates and classification of those individual rates into the number of rate categories specified by c This is only a work around for
15. well for very long alignments but performance is extremely hardware dependent It currently appears to scale best on AMD shared memory nodes as well as the recent multi core platforms and the new SUN x4600 systems while scalability is significantly worse on current Intel architectures It also scales well on the SGI Altix which is very large shared memory supercomputer architecture Even for short alignments 1 900 taxa 1 200bp DNA data we observed speedups of around factor 6 5 on an 8 way shared memory Opteron processor on the CIPRES CyberlInfrastructure for Phyligenetic RESearch http www phylo org cluster For a long alignment 125 taxa 20 000 base pairs DNA we observed significant super linear soeedups of around 10 11 on the 8 way CIPRES SMP nodes those are traditional shared memory nodes not multi cores In general the Pthreads version is more efficient i e yields higher parallel efficiency and better speedups than the previous OpenMP based 13 version In addition it is easier to compile because you do not need an OpenMP compiler any more just the Pthreads library which is pretty much available by default on all Linux and MAC based systems The MPI version is for executing really large production runs i e 100 or 1 000 bootstraps on a LINUX cluster You can also perform multiple inferences on larger datasets in parallel to find a best known ML tree for your dataset Finally the novel rapid BS algorithm and the associ
16. worthwhile to use the a option for a couple of runs to see how the program behaves on completely random starting trees This is where the option as well as the parallel MPI version raxm1HPC MPI come into play So to execute a multiple inference on the original alignment on a single processor just specify raxmlHPC f d m GTRMIX s ex_al 10 n MultipleOriginal and RAxML will do the rest for you Note that specifying m GTRCAT in combination with is not a good idea because you will probably want to compare the trees inferred under GTRCAT based on their likelihood values and will have to compute the likelihood of the final trees under GTRGAMMA anyway Thus you should better use m GTRMIX for those analyses If you have a PC cluster available you would specify raxmlHPC MPI f d m GTRMIX s ex_al 100 n MultipleOriginal preceded by the respective MPI run time commands e g mpiexec or mpirun depending on your local installation please check with your local computer scientist It is important to note that you should specify the execution of one more process than CPUs available e g you have 8 CPUs start 9 MPI processes since one of those is just the master process which collects data and issues jobs to the worker processes and does not produce significant computational load 5 2 4 Bootstrapping with RAxML To carry out a multiple non parametric bootstrap with the sequential version of RAxML just type raxmlHPC f d m GTRCA
17. C runs on the CIPRES project cluster http 8ball sdsc edu 8889 cipres web Bootstrap do In addition RAxML is currently being integrated into the Distributed European Infrastructure for Su percomputing Applications system http www deisa org but am not directly involved in this and only provide some occasional support The RAxML DEISA integration is currently supposed to be in the beta testing phase 1 5 Citing RAxML If you use RAxML please always cite the following paper Alexandros Stamatakis RAXML VI HPC Maximum Likelinood based Phylogenetic Analyses with Thousands of Taxa and Mixed Models Bioinfor matics 22 21 2688 2690 2006 6 In additon when using the Web Servers or the rapid Bootstrapping algorithm please also cite Alexandros Stamatakis Paul Hoover and Jacques Rougemont A Rapid Bootstrap Algorithm for the RAxML Web Servers to be published In case you use the parallel Pthreads based version please also cite Michael Ott Jaroslaw Zola Srinivas Aluru Alexandros Stamatakis Large scale Maximum Likelihood based Phylogenetic Analysis on the IBM BlueGene L in Proceedings of ACM IEEE Supercomputing conference 2007 10 While this paper does not really describe the Pthreads based version information on Pthreads https computing 11n1 gov tutorials pthreads manuscript in preparation an analogous parallelization scheme is used which is more efficient than the previous OpenMP based shared memory
18. D A Increasing the Efficiency of Searches for the Maximum Likelihood Tree in a Phyloge netic Analysis of up to 150 Nucleotide Sequences Systematic Biology 56 2007 988 1010 12 Stamatakis A Auch A Meier Kolthoff J Goeker M AxPcoords amp parallel AxParafit statistical co phylogenetic analyses on thousands of taxa BMC Bioinformatics 8 2007 405 13 Stamatakis A Ott M Ludwig T Raxml omp An efficient program for phylogenetic inference on smps In Proc of PaCTO5 2005 288 302 14 Stamatakis A Phylogenetic models of rate heterogeneity A high performance computing perspec tive In Proc of IPDPS2006 Rhodos Greece 2006 15 Minh B Vinh L Haeseler A Schmidt H piqpnni parallel reconstruction of large maximum likelinood phylogenies Bioinformatics 2005 16 Ripplinger J Sullivan J Does Choice in Model Selection Affect Maximum Likelihood Analysis Systematic Biology 57 2008 76 85 17 Tavar S Some Probabilistic and Statistical Problems in the Analysis of DNA Sequences Some Mathematical Questions in Biology DNA Sequence Analysis 17 1986 18 Yang Z Maximum likelihood phylogenetic estimation from dna sequences with variable rates over sites J Mol Evol 39 1994 306 314 19 Yang Z Among site rate variation and its impact on phylogenetic analyses Trends Ecol Evol 11 1996 367 372 20 Dunn C W Hejnol A Matus D Q Pang K Browne W E Smith S A Sea
19. Mouse Rat Don t leave spaces between the taxon names in the list If there is more than one outgroup a check for monophyly will be performed If the outgroups are not monophyletic the tree will be rooted at the first outgroup in the list and a respective warning will be printed Example raxmlHPC s alg m GTRGAMMA o Rat Mouse n TEST P Specify a random number seed for the parsimony inferences This allows you and others to reproduce your results reproducible verifiable experiments and will help me debug the program This option HAS NO EFFECT in the parallel MPI version Example raxmlHPC s alg m GTRGAMMA p 12345 n TEST P proteinModel Specify the file name of an external AA substitution model The file proteinModel must contain a total of 420 floating point number entries in plain ASCII text which can be separated by any kind of whitespaces tabs spaces linebreaks etc The first 400 entries are the substitution rates of the 20 by 20 AA matrix stored and interpreted in row first order i e the first 20 entries correspond to the first row of the matrix and the last 20 entries entries 401 420 are the base frequencies It is important that the base frequencies sum to 1 0 e since even relatively small deviations might cause numerical instability of AA models The 400 entries of the 20 by 20 matrix must be symmetric the program will check if this is the case The entries on the diagonal matrix will be disregarded since they
20. P Invar Ziheng Yang kindly provided some additional references that refer to this problem 37 38 39 40 25 He also addresses the issue in his recently published book on Computational Molecular Evolution Ox ford University Press 2006 quote from pages 113 114 The model is known as l G and has been widely used This model is somewhat pathological as the gamma distribution with alpha 1 already allows for sites with very low rates as a result adding a proportion of invariable sites creates a strong correlation between p0 and alpha making it impossible to estimate both parameters reliably 39 40 25 Another drawback of the model is that the estimate of p0 is very sensitive to the number and divergences of the sequences included in the data The proportion pO is never larger than the observed proportion of constant sites with the addition of more and divergent sequences the proportion of constant sites drops and the estimate of pO tends to go down as well In any case have so far not encountered any difficulties with reviews for the few real phylogenetic analyses 41 42 have published with collegues from Biology when we used GTR I instead of the more widely spread GTR I l Q Why does RAxML only implement GTR based models of nucleotide substitution For each distinct model of nucleotide substitution RAxML uses a separate highly optimized set of likeli hood functions The idea behind this is that GTR is the most common and gene
21. RGAMMA r constr n TEST s sequenceFileName Specify the name of the alignment data file which must be in relaxed PHYLIP format Relaxed means that you don t have to worry if the sequence file is interleaved or sequential and that the taxon names are too long t userStartingTree Specifies a user starting tree file name which must be in Newick format Branch lengths of that tree will be ignored Note that you can also specify a non comprehensive not containing all taxa in the alignment starting tree now This might be useful if newly aligned sequenced taxa have been added to your alignment Initially taxa will be added to the tree using the MP criterion The comprehensive tree will then be optimized under ML Example raxmlHPC s alg m GTRGAMMA t tree n TEST T PTHREADS VERSIONONLY Specify the number of threads you want to run MaKeSUf6 16 S t T16atiost This option is set to 0 by default the Pthreads version will produce an error if you do not set T to at least 2 Example raxmlHPC PTHREADS T 4 s alg m GTRGAMMA n TEST u Specify the number of multiple BS searches per replicate to obtain better ML trees for each replicate By default only one ML search per BS replicate is conducted This option only works with standard bootstrapping via b not the fast one via x V Displays version information w workingDirectory Name of the working directory where RAxML shall write its output files to X S
22. RGAMMA despite the fact that T is a beautiful Greek letter The main idea behind GTRCAT is to allow for integration of rate heterogeneity into phylogenetic analyses at a significantly lower computational cost about 4 times faster and memory consumption 4 times lower Essentially GTRCAT represents a rather un mathematical quick amp dirty approach to rapidly navigate into portions of the tree space where the trees score well under GTRGAMMA However due to the way individual rates are optimized and assigned to rate categories in GTRCAT for details on this please read the paper 14 the likelihood values computed by GTRCAT are completely meaningless This means You will probably obtain a biased assessment of trees This is the reason why GTRCAT is called approximation instead of model The same applies to the CAT approximation when used with AA data Finally note that in the few real world phylogenetic studies have worked on so far in collaboration with Biologists we never received nasty reviewer comments for using the CAT approximation A very recent phylogenetic analysis with RAxML in Nature also used the CAT approximation 20 3 Installation Compilers Platforms RAxML 7 0 3 can be download at icwww epf1 ch stamatak as open source code under the GNU General Public Licence GPL To install RAXML 7 0 3 download the RAxML 7 0 3 tar gz archive and uncompress it This version comes in three flavors 1 raxmlHPC just the s
23. T Meier H Raxml iii A fast program for maximum likelinood based inference of large phylogenetic trees Bioinformatics 21 2005 456 463 24 47 Robinson D F Foulds L R Comparison of weighted labelled trees Lecture Notes in Mathematics 748 1979 119 126 48 Robinson D F Foulds L R Comparison of Phylogenetic Trees Mathematical Biosciences 53 1981 131 147 25
24. T s ex_al 100 b 12345 n MultipleBootstrap You have to specify a random number seed after b for the random number generator This will allow you to generate reproducible results Note that we can use GTRCAT here if we do not want to compare final trees based on ML scores or need bootstrapped trees with branch lengths To do a parallel bootstrap type raxmlHPC MPI f d m GTRCAT s ex_al 100 b 12345 n MultipleBootstrap once again preceded by the appropriate MPI execution command Note that despite the fact that you specified a random number seed the results of a parallel bootstrap are not reproducible 5 2 5 Obtaining Confidence Values Suppose that you have executed 200 inferences on the original alignment and 1 000 bootstrap runs You can now use the RAxML f b option to draw the information from the 1 000 bootstrapped topologies onto some tree and obtain a topology with support values From my point of view the most reasonable thing to do is to draw them on the best scoring ML tree from those 200 runs Suppose that the best scoring tree was found in run number 99 and the respective tree file is called RAxML_result MultipleOriginal RUN 99 If you have executed more than one bootstrap runs with the sequential version of RAxML on distinct computers i e 10 runs with 100 bootstraps on 10 machines you will first have to concatenate the boot strap files If your bootstrap result files are called e g RAxML_bootstrap MultipleBootstrap 0
25. The RAxML 7 0 3 Manual Alexandros Stamatakis The Exelixis Lab Teaching amp Research Unit Bioinformatics Department of Computer Science Ludwig Maximilians Universitat M nchen stamatakis bio ifi lmu de 1 About RAxML RAxML Randomized Axelerated Maximum Likelihood is a program for sequential and parallel Maximum Likelihood 1 based inference of large phylogenetic trees It has originally been derived from fastDNAm1 which in turn was derived from Joe Felsentein s dnam1 which is part of the PHYLIP 2 package 1 1 What s new in version 7 0 3 e Several bugs in version 7 0 0 fixed e Added code for automatic thread pinning 3 enforcment of thread affinity see Section 3 2 under LINUX e Option to read in external protein substitution matrix model e Option to compute Expected Likelihood Weights Statistics ELW 4 e Added option N as an alternative to to avoid problems caused by with certain job submission problems 1 2 RAxML 7 0 3 In addition to the sequential version RAXML offers two ways to exploit parallelism fine grained parallelism that can be exploited on shared memory machines or multi core architectures and coarse grained paral lelism that can be exploited on Linux clusters The current version of RAXML is a highly optimized program which handles DNA and AA alignments under various models of substitution and several distinct methods of rate heterogeneity In addition it implements a significantly improve
26. You don t need to run it with the default setting of c 25 since you already have that data such that you can continue with raxmlHPC f d c 40 m GTRMIX s ex_al t RAxML_parsimonyTree STO n C40_0 raxmlHPC f d c 40 m GTRMIX s ex_al t RAxML_parsimonyTree ST4 n C40_4 and so on and so forth Since the GTRCAT approximation is still a new concept little is known about the appropriate setting for c 25 However empirically c 25 worked best on 19 real world alignments So testing up to c 55 should usually be sufficient except if you notice a tendency for final GTRGAMMA likelihood values to further improve with increasing rate category number Thus the assessment of the good c setting should once again be based on the final GTRGAMMA likeli hood values If you don t have the time or computational power to determine both good c and i settings you should rather stick to determining i since it has shown to have a greater impact on the final results Also note that increasing the number of distinct rate categories has a negative impact on execution times Finally if the runs with the automatic determination of the rearrangement settings from Section 5 2 1 have yielded the best results you should then use exactly the same rearrangement settings for each series of experiments to determine a good c setting The automatically determined rearrangement settings can be retrieved from file RAxML_info AI_O RAxML_info AI_4
27. airs We also inferred trees for a dataset of 250 taxa and about 500 000 base pairs the to the best of the author s knowledge largest dataset analyzed under ML to date on 1 024 processors of a Blue Gene supercomputer 10 The BlueGene version is a specialized unreleased RAXML version available upon request but the concepts developed in this paper are currently being integrated into the standard RAxML release Finally RAxML despite being developed for handling large datasets also does fine on smaller to medium sized datasets see 11 for a respective performance study on datasets up to 150 taxa 1 3 RAxML Community Contributions Several people have contributed to make RAxML easier to use and make it available on more platforms would like to express my gratitude to all of them My colleague Frank Kauff now at University of Kaiserslautern fkauff rhrk uni kl de previously at Duke University has written a cool biopython wrapper called PYRAXML2 This is a script that reads NEXUS style data files and prepares the necessary input files and command line options for RAxML You can download the Beta version at http www 1lutzonilab net downloads My colleague Olaf Bininda Emonds olaf bininda uni jena de has written a perl script that provides a wrapper around RAxML to easily analyze a set of data files according to a common set of search criteria It also organizes the RAxML output into a set of subdirectories You can download it at
28. algorithm of version 2 1 3 6 this is essentially just for backward compatibility f p performs just pure stepwise MP addition of new sequences to an incomplete starting tree Example raxmlHPC f p t ref s alg m GTRCAT n TEST f s option can be used to split a multi gene alignment into individual genes provided a model file with q This might be useful to select best fitting models for individual partitions of an AA multi gene alignment or to infer per partition trees in order to analyze tree compatibility Example raxmlHPC f s q part s alg m GTRCAT n TEST f t will perform N randomized tree searches that always start from one fixed starting tree Example raxmlHPC f t t ref 100 s alg m GTRCAT n TEST f w will perform an ELW test 4 on a bunch of input trees passed via z You will also need to specify a BS seed via b and and the number of replicates you want to compute via N This test does obvisouly not work under the CAT approximation Example raxmlHPC f w z trees 100 b 12345 s alg m GTRGAMMA n TEST g groupingFileName This option allows you to specify an incomplete or comprehensive multifurcating constraint tree for the RAxML search in NEWICK format Initially multifurcations are resolved randomly If the tree is incomplete does not contain all taxa the remaining taxa are added by using the MP criterion Once a comprehen sive containing all taxa bifurcating tree is computed it is f
29. ally cause some error messages during compila tion of RAxML If this happens please send an email to stamatakis bio ifi 1lmu deand ottmi in tum de 4 The RAxML Formats Options amp Output Files 4 1 Input Alignment amp Input Tree Formats The input alignment format of RAXML is relaxed interleaved or sequential PHYLIP Relaxed means that sequence names can be of variable length between 1 up to 100 characters If you need longer taxon names you can adapt the constant define nmlngth 100 in file axml h appropriately Moreover RAxML should be less sensitive with respect to the formatting tabs insets etc of interleaved PHYLIP files The input tree format is Newick see http evolution genetics washington edu phylip newicktree htm1 the RAXML input trees must not be comprehensive i e need not contain all taxa 4 2 Alignment Error Checking recently noticed that a lot of alignments should be checked for the following errors insufficiencies before running an analysis with RAxML or any other phylogenetic inference program RAXML will now analyze the alignment and check for the following errors Identical Sequence name s appearing multiple times in an alignment this can easily happen when you export a standard PHYLIP file from some tool which truncates the sequence names to 8 or 10 char acters Identical Sequence s that have different names but are exactly identical This mostly happens when you excluded some hard to align alig
30. ated ML search have also been parallelized with MPI The best hardware to run RAxML on is currently the AMD Opteron 13 architecture 3 2 Processor Affinity and Thread Pinning with the Pthreads Version An important aspect if you want to use the Pthreads version of the program is to find out how your operating system platform handles processor affinity of threads Within the shared memory or multi core context processor affinity means that if you run e g 4 threads on a 4 way CPU or 4 cores each individual thread should always run on the same CPU i e threadO on CPUO threadi on CPU1 etc This is important for efficiency since cache entries can be continuously re used if a thread which works on the same part of the shared memory space remains on the same CPU If threads are moved around e g thread0 is initially executed on CPUO but then on CPU4 etc the cache memory of the CPU will have to be re filled every time a thread is moved With processor affinity enabled performance improvements of 5 have been measured on sufficiently large and thus memory intensive datasets On multi core systems the analysis of memory access patterns and cache congestion is more com plicated we are currently looking at this and will provide additional information and hopefully appropraite solutions soon Version 7 0 3 now contains a function that will automatically pin threads to CPUs i e enforce thread affinity under LINUX UNIX This function might occasion
31. d version run time improvement of factor 2 5 of the fast rapid hill climbing algorithm 5 compared to the algorithm described in 6 At the same time these new heuristics yield qualitatively comparable results In addition to this it also offers a novel unpublished rapid Bootstrapping 7 algorithm that is faster by at least one order of magnitude than all other current implementations RAxML 2 2 3 GARLI 8 PHYML 9 Once again the results obtained by the rapid bootstrapping algorithm are qualitatively comparable to those obtained via the standard RAxML BS algorithm and more importantly the deviations in support values be tween the rapid and the standard RAxML BS algorithm are smaller than those induced by using a different search strategy e g GARLI or PHYML This rapid BS search can be combined with a rapid ML search on the original alignment and thus allows users to conduct a full ML analysis within one single program run Some data structures have been changed and functions re written Those technical changes yield an additional run time improvement of around 5 1 Exelixis is the Greek word for evolution The program has been developed to be able to handle extremely large datasets such as a single gene 25 000 taxon alignment of protobacteria length approximately 1 500 base pairs run time on a single CPU 13 5 days memory consumption 1 5GB or a large multi gene alignment of 2 100 mammals with a length of over 50 000 base p
32. elcodon2 2 500 3 DNA genelcodon3 3 500 3 DNA gene2 501 1000 If you only need a distinct model for the 3rd codon position you can write DNA genetcodoniandcodon2 1 500 3 2 500 3 DNA genelcodon3 3 500 3 DNA gene2 501 1000 As already mentioned for AA data you must specify the transition matrices for each partition JTT genet 1 500 WAGF gene2 501 800 WAG gene3 801 1000 The AA substitution model must be the first entry in each line and must be separated by a comma from the gene name just like the DNA token above You can not assign different models of rate heterogeneity to different partitions i e it will be either CAT GAMMA GAMMAT etc for all partitions as specified with m Finally if you have a concatenated DNA and AA alignment with DNA data at positions 1 500 and AA data at 501 1 000 with the WAG model the partition file should look as follows DNA genel 1 500 WAG gene2 501 1000 Example raxmlHPC s alg m GTRGAMMA q part n TEST r constraintFileName This option allows you to pass a binary bifurcating constraint backbone tree in NEWICK format to RAxML Note that using this option only makes sense if this tree contains less taxa than the input alignment The remaining taxa will initially be added by using the MP criterion Once a comprehensive tree with all taxa has been obtained it will be optimized under ML respecting the restrictions of the constraint tree Example raxmlHPC s alg m GT
33. es are categorized using the 4 discrete GAMMA rates that are assigned to sites following a formula by Yang Evaluation of the final tree topology will be conducted under specified AA matrix GAMMA This is mostly for experimental purposes m PROTGAMMAImatrixName F Same as PROTGAMMAmatrixName F but with estimate of proportion of invariable sites 11 m PROTMIXImatrixName F Same as PROTMIXmatrixName F but with estimate of proportion of invari able sites m PROTCAT_GAMMAImatrixName F Same as PROTCAT_GAMMAmatrixName F but with estimate of pro portion of invariable sites M Switch on estimation of individual per partition branch lengths Only has effect when used in combination with q and an alignment partition file Branch lengths for individual partitions will be printed to separate files A weighted average of the branch lengths is also computed by using the respective partition lengths number of columns per partition Note that this does not take into account the gappyness of partitions but am currently not sure how to solve this problem By default the M option is turned off for partitioned analyses i e RAxML will compute a joined branch length estimate Example raxmlHPC s alg m GTRGAMMA q part M n TEST n outputFileName Specify the name of this run according to which the various output files will be named o outgroupName s Specify the name names of the outgroup taxa e g o Mouse or o
34. f Amino Acid Substitution in Proteins Encoded by Mitochondrial DNA Journal of Molecular Evolution 42 1996 459 468 30 Whelan S Goldman N A General Empirical Model of Protein Evolution Derived from Multiple Protein Families Using a Maximum Likelihood Approach Molecular Biology and Evolution 18 2001 691 699 31 Dimmic M Rest J Mindell D Goldstein R rtREV An Amino Acid Substitution Matrix for Inference of Retrovirus and Reverse Transcriptase Phylogeny Journal of Molecular Evolution 55 2002 65 73 32 Adachi J Waddell P Martin W Hasegawa M Plastid Genome Phylogeny and a Model of Amino Acid Substitution for Proteins Encoded by Chloroplast DNA Journal of Molecular Evolution 50 2000 348 358 33 Mueller T Vingron M Modeling Amino Acid Replacement Journal of Computational Biology 7 2000 761 776 34 Henikoff S Henikoff J Amino Acid Substitution Matrices from Protein Blocks Proceedings of the National Academy of Sciences of the United States of America 89 1992 10915 10919 35 Yang Z Synonymous and Nonsynonymous Rate Variation in Nuclear Genes of Mammals Journal of Molecular Evolution 46 1998 409 418 36 Gu X Maximum likelihood estimation of the heterogeneity of substitution rate among nucleotide sites Molecular Biology and Evolution 12 1995 546 557 37 Ren F Tanaka H Yang Z An Empirical Examination of the Utility of Codon Substitution Models in Phylogeny Reconstruc
35. fer trees with RAxML and to analyze really large datasets several genes or more than 1 000 taxa or to conduct a large number of BS replicates is to use the novel rapid BS algorithm and combine it with an ML search RAxML will then conduct a full ML analysis i e a certain number of BS replicates and a search for a best scoring ML tree on the original alignment To just do a BS search you would type raxmlHPC x 12345 p 12345 100 m GTRGAMMA s ex_al n TEST Note that the rapid BS algorithm will override the choice of GTRGAMMA and always use the GTR CAT approximation for efficiency Thus whether you specify m GTRGAMMA m GTRCAT m GTRGAMMAI the re sult will always be the same Note that here added the p option to pass a random number seed for MP starting tree computations such that the results of the analysis will always be the same would like to encourage users to do so as well because this will allow me to reconstruct potential bugs more easily Now if you want to run a full analysis i e BS and ML search type raxmlHPC f a x 12345 p 12345 100 m GTRGAMMA s ex_al n TEST This will first conduct a BS search and once that is done a search for the best scoring ML tree Such a program run will return the bootstrapped trees RAxML_bootstrap TEST the best scoring ML tree RAxML_bestTree TEST and the BS support values drawn on the best scoring tree RAxML_bipartitions TEST Here the model choice via m plays a role f
36. http www personal uni jena de b6biol12 ProgramsMain html James Munro munroj01 student ucr edu at UCR has put up a web site that provides a guide for in stalling RAxML on MACs http hymenoptera ucr edu index php option com_content amp task viewkid 62 amp Itemid s8 Dave Carmean carmean sfu ca at Simon Fraser University has kindly assembled a RAxML executable for MACs and put up a web site entitled Installing and running RAxML on a Mac in less than a minute http www sfu ca biology staff dc raxml1 Graham Jones http www sightsynthesis co uk has provided invaluable help by contributing the Windows executable of RAxML Finally Andreas Tille at the Robert Koch Institute tillea rki de has pushed forward the integration of RAxML and AxParafit another open source Bioinformatics code have developped see 12 into the Debian med package for details on this project see http www debian org devel debian med 1 4 RAxML Web Servers Together with Jacques Rougemont formerly at the Vital IT Unit of the Swiss Institute of Bioinformatics now at EPFL jacques rougemont epfl ch and Paul Hoover at the San Diego Supercomputer Center phoover sdsc edu we have developed two RAxML Web Servers that offer the novel rapid RAxML Boot strapping algorithm and thorough ML searches on the original alignments The one in Switzerland is located at the Vital IT unit of the SIB http phylobench vital it ch raxml bb and the one at SDS
37. ined by different RAxML versions especially those prior to version 2 1 0 should not be directly compared with each other either The same holds for compar isons of likelihood values between RAxML VI HPC v2 2 3 and RAxML 7 0 3 This is due to frequent code and data structure changes in the likelihood function implementation and model parameter optimization procedures Thus if you want to compare topologies obtained by distinct ML programs make sure that you optimize branch lengths and model parameters of final topologies with one and the same program This can be done by either using the respective RAxML option f e or e g the corresponding option in PHYML 9 In theory all ML programs implement the same mathematical function and should thus yield the same likelinood score for a fixed model and a given tree topology However if we try to implement a numerical function on a finite machine we will unavoidably obtain rounding errors Even if we change the sequence or if it is changed by the compiler of some operations applied to floating point or double precision arithmetics in our computer we will probably get different results In my experiments have observed differences among final likelihood values between GARLI IQPNNI PHYML RAxML every program showed a different value You can also experiment with this by removing the gcc optimization flag 03 in the RAXML Makfile This will yield much slower code that is in theory mathematically eq
38. inted when f a x have been specified at the end of the analysis the program will draw the BS support values on the best tree found during the ML search RAXML_reducedList exampleRun If you used 1 or L this file will contain clustering information in the following format taxi tax2 tax3 tax4 tax10 tax9 taxil where the first entry in each line is the taxon name of the respective representative sequence of a cluster while the remaining ones after are the taxa that have been removed via clustering RAxML_bipartitionFrequencies exampleRun Contains the pair wise bipartition frequencies of all trees contained in files passed via t and z when the f m option has been used RAxML_perSiteLLs exampleRun Contains the per site log likelihood scores in Treepuzzle format for us age with CONSEL 22 This file is only printed when f g is specified RAxML_bestTree exampleRun Contains the best scoring ML tree of a thorough ML analysis in conjunc tion with a rapid BS analysis i e when options x 12345 f a are used 5 How to set up and run a typical Analysis This is a HOW TO which describes how RAxML should best be used for a real world biological analysis given an example alignment named ex_al Section 5 1 covers the easy fully automatic fast way to run it using the novel rapid BS algorithm while Section 5 2 describes the hard more compute intensive and more thorough way 5 1 The Easy amp Fast Way The easy and fast way to in
39. istic For example only few methods exist that incorporate the generation of gaps in simulated alignments Since the model according to which the sequences are generated on the true tree is pre defined we are actually assuming that ML exactly models the true evolutionary process while in reality we simply don t know how sequences evolved The above simplifications lead to perfect alignment data without gaps that evolved exactly according to a pre defined model and thus exhibits a very strong phylogenetic signal in contrast to real data In addition the given true tree must not necessarily be the Maximum Likelihood tree This difference manifests itself in substantially different behaviors of search algorithms on real and simulated data Typically search algorithms execute significantly less factor 5 10 topological moves on simulated data until convergence as opposed to real data i e the number of successful Nearest Neighbor Interchanges NNIs or subtree rearrangements is lower Moreover in several cases the likelihood of trees found by RAxML on simulated data was better than that of the true tree Another important observation is that program performance can be inverted by simulated data Thus a program that yields good Robinson Foulds distances 47 48 on simulated data can in fact perform much worse on real data than a program that does not perform well on simulated data If one is willing to really accept ML as inference crite
40. it has been observed that this might sometimes yield topologies of distinct local likelihood maxima which better correspond to empirical expectations Example raxmlHPC d s alg m GTRGAMMA n TEST e likelihoodEpsilon This allows you to specify up to which likelihood difference i e the model parameters will be optimized when you use either the GTRGAMMA or GTRMIX models or when you just evaluate final trees with the f e option This has shown to be useful to quickly evaluate the likelihood of a bunch of large final trees of more than 1 000 taxa because it will run much faster typically use e g e 1 0 or e 2 0 in order to rapidly compare distinct final tree topologies based on their likelihood values Note that topology dependent likelihood differences are typically far larger than 1 0 or 2 0 log likelihood units The default setting is 0 1 log likelihood units which proves to be sufficient in most practical cases Example raxmlHPC e 0 00001 s alg m GTRGAMMA n TEST sE Used to specify an exclude file name that contains a specification of alignment positions you wish to ex clude from your analysis The format is similar to Nexus the file shall contain entries like 100 200 300 400 to exclude e g all columns between positions 100 and 200 as well as all columns between positions 300 and 400 Note that the bounds i e positions 100 200 300 and 400 will also be excluded To exclude a single column write e g 100 100 This
41. nces University of Washington Seattle 3 Ott M Klug T Weidendorfer J Trinitis C autopin Automated Optimization of Thread to Core Pinning on Multicore Systems In Proceedings of 1st Workshop on Programmability Issues for Multi Core Computers MULTIPROG 2008 4 Strimmer K Rambaut A Inferring confidence sets of possibly misspecified gene trees Proc R Soc Lond B 269 2002 137 142 5 Stamatakis A Blagojevic F Nikolopoulos D Antonopoulos C Exploring New Search Algorithms and Hardware for Phylogenetics RAxML Meets the IBM Cell The Journal of VLSI Signal Processing 48 2007 271 286 6 Stamatakis A RAXML VI HPC maximum likelihood based phylogenetic analyses with thousands of taxa and mixed models Bioinformatics 22 2006 btl446 7 Felsenstein J Confidence Limits on Phylogenies An Approach Using the Bootstrap Evolution 39 1985 783 791 22 8 Zwickl D Genetic Algorithm Approaches for the Phylogenetic Analysis of Large Biological Sequence Datasets under the Maximum Likelihood Criterion PhD thesis University of Texas at Austin 2006 9 Guindon S Gascuel O A simple fast and accurate algorithm to estimate large phylogenies by maximum likelihood Syst Biol 52 2003 696 704 10 Ott M Zola J Aluru S Stamatakis A Large scale Maximum Likelihood based Phylogenetic Analysis on the IBM BlueGene L In ACM IEEE Supercomputing conference 2007 2007 11 Morrison
42. nment regions from your alignment Undetermined Column s that contain only ambiguous characters that will be treated as missing data i e columns that entirely consist of X for AA data and N 0 X for DNA data Undetermined Sequence s that contain only ambiguous characters see above that will be treated as missing data In case that RAxML detects Identical Sequences and or Undetermined Columns and was executed e g with n alignmentName it will automatically write an alignment file called alignmentName reduced with Identical Sequences and or Undetermined Columns removed If this is detected for a multiple model analysis a respective model file modelFileName reduced will also be written In case RAxML encounters identical sequence names or undetermined sequences it will exit with an error and you will have to fix your alignment 4 3 Program Options raxmlHPC MPI PTHREADS s sequenceFileName n outputFileName m substitutionModel a weightFileName b bootstrapRandomNumberSeed c number0fCategories d e likelihoodEpsilon E excludeFileName f alblcildlelglhliljlminlolplsItlw g groupingFileName h i initialRearrangementSetting j k 1 sequenceSimilarityThreshold L sequenceSimilarityThreshold am M o outGroupName1 outGroupName2 p parsimonyRandomSeed P proteinModel q multipleModelFileName r binaryConstraintTree t userStartingTree T number0fThreads
43. or the ML search If you specify e g m GTRCAT the ML search will still be conducted under GTRGAMMA So in general the only thing that matters is whether you want to use GTRGAMMA or GTRGAMMAT to include an estimate of the proportion of invariable sites Finally note that by increasing the number of BS replicates via you will also make the ML search more thorough since for ML optimization every 5th BS tree is used as a Starting point to search for ML trees From what have observed so far this new ML search algorithm yielded better trees than what is obtained via 20 standard ML searches on distinct starting trees for all datasets with lt 1 000 sequences For larger datasets it might be worthwhile to conduct an additional ML search as described in Section 5 2 3 just to be sure 5 2 The Hard amp Slow Way Despite the observation that the default parameters and the rapid BS and ML algorithm described above work well in most practical cases a good thing to do is to adapt the program parameters to your alignment This refers to a good setting for the rate categories of m GTRCAT and the initial rearrangement setting If you use mixed models you should add q modelFileName to all of the following commands 5 2 1 Getting the Initial Rearrangement Setting right If you don t specify an initial rearrangement setting with the i option the program will automatically deter mine a good setting based upon the randomized MP starting tree I
44. orough ML search on the original alignment We introduced N as an alternative to since the special character seems to sometimes cause problems with certain batch job submission systems In combination with j this will generate numberOfRuns bootstrapped alignment files Example raxmlHPC s alg n TEST m GTRGAMMA 20 4 4 Output Files Depending on the search parameter settings RAXML will write a number of output files The files a run named n exampleRun will write are listed below RAXxML_log exampleRun A file that prints out the time likelihood value of the current tree and number of the checkpoint file if the use of checkpoints has been specified after each iteration of the search algorithm In the last line it also contains the final likelihood value of the final tree topology after thorough model optimization but only if m GTRMIX or m GTRGAMMA have been used This file is not written if multiple bootstraps are executed i e and b have been specified In case of a multiple inference on the original alignment option the Log Files are numbered accordingly RAxML_result exampleRun Contains the final tree topology of the current run This file is also written after each iteration of the search algorithm such that you can restart your run with t in case your computer crashed This file is not written if multiple bootstraps are executed i e and b have been specified RAXML_info exampleRun contains information
45. pecify an integer number random seed and turn on rapid bootstrapping This will invoke the novel rapid bootstrapping algorithm Example raxmlHPC x 12345 100 m GTRCAT s alg n TEST y If you want to only compute a randomized parsimony starting tree with RAxML and not execute an ML analysis of the tree specify y The program will exit after computation of the starting tree This option can be useful if you want to assess the impact of randomized MP and Neighbor Joining starting trees on your search algorithm They can also be used e g as starting trees for Derrick Zwickl s GARLI program for ML inferences which needs comparatively good starting trees to work well above approximately 500 taxa z multipleTreesFile Only effective in combination with the b f h f m f n options This file should contain a number of trees in NEWICK format The file should contain one tree per line without blank lines between trees For example you can directly read in a RAxML bootstrap result file with z 14 N numberOfRuns Specifies the number of alternative runs on distinct starting trees e g if 10 or N 10 is specified RAxML will compute 10 distinct ML trees starting from 10 distinct randomized maximum parsimony starting trees In combination with the b option this will invoke a multiple bootstrap analysis In combination with x this will invoke a rapid BS analysis and combined with f a x a rapid BS search and thereafter a th
46. porate rate heterogeneity in order to obtain publishable results Put aside the publish or perish argument there is also strong biological evidence for rate heterogeneity among sites The rationale for being sceptical about P Invar in RAxML is that all three alternatives GTRGAMMA GTRCAT and P Invar represent distinct approaches to incorporate rate heterogeneity Thus in principle they account for the same phenomenon by different mathematical means Also some unpublished concerns have been raised that the usage of P Invar in combination with T can lead to a ping pong effect since a change of P Invar leads to a change in T and vice versa This essentially means that those two parameters i e a and P Invar can not be optimized independently from each other and might cause significant trouble and problems during the model parameter everything except tree topology optimization process In fact already observed this when was implementing P Invar in RAXML on a very small AA dataset Although this has never been properly documented several well known researchers in phylogenetics share this opinion Arndt v Haeseler Ziheng Yang quote from an recent email in 2008 regarding this part of the RAxML manual entirely agree with your criticism of the Pinv Gamma model even though as you said it is very commonly used Korbinian Strimmer personal communications The following paper 36 touches this problem of dependency between a and
47. ral model for real world DNA analysis Thus it is better to efficiently implement and optimize this model instead of offering a plethora of distinct models which are only special cases of GTR but are programmed in a generic and thus inefficient way My personal view is that using a simpler model than GTR only makes sense with respect to the computational cost i e it is less expensive to compute Programs such as Modeltest 43 propose the usage of a simpler model for a specific alignment if the likelinood of a fixed topology under that simpler model is not significantly worse than that obtained by GTR based on a likelihood ratio test My experience is that GTR always yields a slightly better likelihood than alternative simpler models In addition 20 since RAxML has been designed for the inference of large datasets the danger of over parameterizing such an analysis is comparatively low Provided these arguments the design decision was taken to rather implement the most general model efficiently than to provide many inefficient generic implementations of models that are just special cases of GTR Finally the design philosophy of RAxML is based upon the observation that a more thorough topological search has a greater impact on final tree quality than modeling details Thus the efficient implementation of a rapid search mechanisms is considered to be more important than model details Note that Derrick Zwickl has independently adapted the same strateg
48. rion on real data one must also be willing to assume that the tree with the best likelinood score is the tree that is closest to the true tree My personal conclusion is that there is a strong need to improve simulated data generation and method ology In addition the perhaps best way to assess the validity of our tree inference methods consists in an empirical evaluation of new results and insights obtained by real phylogenetic analysis This should be based on the prior knowledge of Biologists about the data and the medical and scientific benefits attained by the computation of phylogenies Q Why am getting weird error messages from the MPI version You probably forgot to specify the option in the command line which must be used for the MPI version to work properly Q When using mixed models can I link the model parameters of distinct partitions to be estimated jointly in a similar as way MrBayes does it Currently not but the implementation of such an option is planned 21 7 Things in Preparation A couple of things are in preparation to be hopefully released within the next 6 months which will further expand the capabilities of RAxML Please be patient with feature requests since do not have anybody to help me with program development e Built in bootstopping convergence criterion e Linking parameter estimation across mixed models e ML based estimate of base frequencies have been promising that for a long time now
49. t will take the starting tree and apply lazy subtree rearrangements with a rearrangement setting of 5 10 15 20 25 The minimum setting that yields the best likelihood improvement on the starting trees will be used as initial rearrangement setting This procedure can have two disadvantages Firstly the initial setting might be very high e g 20 or 25 and the program will slow down considerably Secondly a rearrangement setting that yields a high improvement of likelihood scores on the starting tree might let the program get stuck earlier in some local maximum this behavior could already be observed on a real dataset with about 1 900 taxa Therefore you should run RAxML a couple of times the more the better with the automatic determi nation of the rearrangement setting and with a pre defined value of 10 which proved to be sufficiently large and efficient in many practical cases In the example below we will do this based on 5 fixed starting trees So let s first generate a couple of randomized MP starting trees Note that in RAxML VI HPC 2 2 3 you also always have to specify a substitution model regardless of whether you only want to compute an MP starting tree with the y option raxmlHPC y s ex_al m GTRCAT n STO raxmlHPC y s ex_al m GTRCAT n ST4 Then infer the ML trees for those starting trees using a fixed setting i 10 raxmlHPC f d i 10 m GTRMIX s ex_al t RAxML_parsimonyTree STO n FIO raxmlHPC f d i 10 m GTRMIX
50. tandard sequential version compile it with gcc by typing make f Makefile gcc for LINUX and MAC 2 raxm1HPC PTHREADS the Pthreads parallelized version of RAxML which is intended for shared memory and multi core architectures It is compiled with the gcc compiler by typing make f Makefile PTHREADS or make f Makefile PTHREADS MAC on MACs 3 raxmlHPC MPI the MPI parallelized version for all types of clusters to perform parallel bootstraps rapid parallel bootstraps or multiple inferences on the original alignment compile with the mpicc MPI compiler by typing make f Makefile MPI Other compilers It might make sense to use the now much improved Intel compiler icc instead of gcc on some systems The icc version 10 0 have on my laptop produces 20 30 faster code than gcc IMPORTANT WARNING FOR MPI and PTHREADS VERSIONS If you want to compile the MPI or PTHREADS version of RAxML but have previously compiled the sequential version make sure to remove all object files of the sequential code by typing rm 0 everything needs to be re compiled for MPI and PTHREADS 4 3 1 When to use which Version The use of the sequential version is intended for small to medium datasets and for initial experiments to determine appropriate search parameters However by using the rapid BS algorithm you can conduct a full ML analysis with RAxML on single gene datasets up to 2 000 taxa within 2 3 days on your desktop The Pthreads version will work
51. the GAMMA model of rate heterogeneity so make sure not to compare alternative topologies based on their PROTCAT based likelihood values Therefore you can not use PROTCAT in combination with f e tree evaluation and not in combina tion with multiple analyses on the original alignment N option This is due to the fact that the author assumes that you want to compare trees based on likelihoods if you do a multiple run on the original alignment If you specify e g one of the m PROTCAT models and 10 the program will automatically use the respective PROTMIX model see below m PROTM XmatrixName F This option will make RAxML perform a tree inference search for a good topology under PROTCAT When the analysis is finished RAxML will switch its model to the re spective PROTGAMMA model and evaluate the final tree topology under PROTGAMMA such that it yields stable likelihood values m PROTGAMMAmatrixName F AA matrix specified by matrixName with the T model of rate hetero geneity All free model parameters are estimated by RAxML The GAMMA implementation uses 4 dis crete rate categories which represents an acceptable trade off between speed and accuracy Note that this has been hard coded for performance reasons i e the number of discrete rate categories can not be changed by the user m PROTCAT_GAMMAmatrixName F Inference of the tree under specified AA matrix and site specific evolutionary rates However here rat
52. tion Systematic Biology 54 2005 808 818 38 Minin V Abdo Z Joyce P Sullivan J Performance Based Selection of Likelihood Models for Phylogeny Estimation Systematic Biology 52 2003 674 683 39 Mayrose l Friedman N Pupko T A Gamma mixture model better accounts for among site rate heterogeneity Bioinformatics 21 2005 40 Sullivan J Swofford D Naylor G The Effect of Taxon Sampling on Estimating Rate Heterogeneity Parameters of Maximum Likelinood Models Molecular Biology and Evolution 16 1999 1347 1356 41 Grimm G W Renner S S Stamatakis A Hemleben V A nuclear ribosomal DNA phylogeny of acer inferred with maximum likelihood splits graphs and motif analyses of 606 sequences Evolutionary Bioinformatics Online 2 2006 279 294 42 Gottschling M Stamatakis A Nindl 1 Stockfleth E Alonso A Gissmann L Bravo I G Multi ple evolutionary mechanisms drive papillomavirus diversification Molecular Biology and Evolution 24 2007 1242 1258 43 Posada D Crandall K Modeltest testing the model of dna substitution Bioinformatics 14 1998 817 818 44 Stamatakis A An efficient program for phylogenetic inference using simulated annealing In Proc of IPDPS2005 Denver Colorado USA 2005 45 Stamatakis A Ludwig T Meier H New fast and accurate heuristics for inference of large phyloge netic trees In Proc of IPDPS2004 2004 46 Stamatakis A Ludwig
53. uivalent to the optimized code but will yield slightly different likelinood scores due to re ordered floating point operations My personal opinion is that the topological search number of topologies analyzed is much more im portant than exact likelihood scores to obtain good final ML trees Especially on large trees with more 2As an example for this you might want to implement a dense matrix multiplication on doubles and then re order the instructions than 1 000 sequences the differences in likelihood scores induced by the topology are usually so large that a very rough parameter optimization with an of 1 log likelihood unit i e if the difference between two successive model parameter optimization iterations is lt 1 0 we stop the optimization will already clearly show the differences Note that if you perform a bootstrap analysis you don t need to worry too much about likelinood values anyway since usually you are only interested in the bootstrapped topologies 2 2 The GTRCAT Mystery The GTRCAT approximation is a computational work around for the widely used see 16 for interesting us age statistics General Time Reversible GTR 17 model of nucleotide substitution under the model of rate heterogeneity 18 19 CAT is used in an analogous way to accomodate searches with rate hetero geneity in the AA substitution models There is a paper available 14 which describes what GTRCAT is and why don t like GT
54. urther optimized under ML respecting the given constraints Important If you specify a non comprehensive constraint e g a constraint tree that does not contain all taxa RAxML will assume that the remaining taxa that are not contained in the constraint topology are unconstrained i e these taxa can be placed in any part of the tree As an example consider an align ment with 10 taxa Loach Chicken Human Cow Mouse Whale Seal Carp Rat Frog If for exam ple you would like Loach Chicken Human Cow to be monophyletic you would specify the constraint tree as follows Loach Chicken Human Cow Mouse Whale Seal Carp Rat Frog Moreover if you would like Loach Chicken Human Cow to be monophyletic and in addition Human Cow to be mono phyletic within that clade you could specify Loach Chicken Human Cow Mouse Whale Seal Carp Rat Frog If you specify an incomplete constraint Loach Chicken Human Cow Mouse Whale Seal Carp the two groups Loach Chicken Human Cow and Mouse Whale Seal Carp will be monophyletic while Rat and Frog can end up anywhere in the tree h If you call raxm1HPC h this will print a summary of the program options to your terminal i initialRearrangementSetting This allows you to specify an initial rearrangement setting for the initial phase of the search algorithm If you specify e g i 10 the pruned subtrees will be inserted up to a distance of 10 nodes away from their
55. ver E Rouse G W Obst M Edgecombe G D Sorensen M V Haddock S H D Schmidt Rhaesa A Okusu A Kris tensen R M Wheeler W C Martindale M Q Giribet G Broad phylogenomic sampling improves resolution of the animal tree of life Nature 2008 advance on line publication 21 Schmidt H Strimmer K Vingron M Haeseler A Tree puzzle maximum likelihood phylogenetic analysis using quartets and parallel computing Bioinformatics 18 2002 502 504 22 Shimodaira H Hasegawa M CONSEL for assessing the confidence of phylogenetic tree selection 2001 23 SHIMODAIRA H HASEGAWA M MULTIPLE COMPARISONS OF LOG LIKELIHOODS WITH AP PLICATIONS TO PHYLOGENETIC INFERENC Molecular biology and evolution 16 1999 1114 1116 24 Heyer L Kruglyak S Yooseph S Exploring Expression Data Identification and Analysis of Coex pressed Genes Genome Research 9 1999 1106 1115 25 Yang Z Maximum likelinood estimation of phylogeny from DNA sequences when substitution rates differ over sites 1993 26 Dayhoff M Schwartz R Orcutt B A model of evolutionary change in proteins Atlas of Protein Sequence and Structure 5 1978 345 352 27 Kosiol C Goldman N Different Versions of the Dayhoff Rate Matrix Molecular Biology and Evolution 22 2005 193 199 23 28 Jones D Taylort W Thornton J A new approach to protein fold recognition Nature 358 1992 86 89 29 Adachi J Model o
56. y in his very good GARLI code http www zo utexas edu faculty antisense Garli htm1 based on similar considerations personal communication Q How does RAxML perform compared to other programs RAxML has been compared to other phylogeny programs mainly based on real world biological datasets and best known likelihood values Those analyses can be found in 6 44 45 46 On almost all real datasets RAxML outperforms other current programs with respect to inference times as well as final likeli hood values An exception is Derrick Zwickl s GARLI code which represents a good alternative to RAxML for trees containing less than approximately 1 000 1 500 taxa The main advantages of RAxML with re spect to all other programs are the highly optimized and efficient likelinood functions and the very low mem ory consumption In particular the implementation of the GTRCAT feature allows RAxML to compute huge trees under a realistic approximation of nucleotide substitution which is currently impossible with competing programs due to excessive memory requirements An initial analysis of the large multi gene mammalian dataset under GTRCAT showed promising results Q Why has the performance of RAXML mainly been assessed using real world data Despite the unquestionable need for simulated data and trees to verify and test the performance of current ML algorithms the current methods available for generation of simulated alignments are not very real
Download Pdf Manuals
Related Search