Manual - Cooper Lab Web Page
Contents
1. 1 Replace missing values with median of columns or rows 2 Replace missing values with mean of columns or rows Go to File gt Settings to launch a window that will allow you to specify which of these two techniques to use AutoSOME GUI The AutoSOME GUI is launched using Java Web Start from your browser or via the command line and is used to run AutoSOME as well as to browse and generate publication quality visualizations of the cluster output Figure 1 Figure 1 FB AutoSOME GUI 2 0 build 09 01 10 File Help e 1 INPUT Input and Output pa Input File ai Output Directory 2 RUN F Autodetect Column Labels My Data Has Column Labels 3 OUTPUT Basic Fields Show Input Adjustment Show Load and Filter Input To begin launch a file browser by either pressing the large INPUT button in the AutoSOME Steps panel on the top left or by clicking the browse button adjacent to the input file text box in the Input and Output section Figure 1 A file browser will appear For standard input see Input Format above simply find your file and open it For microarray input optional data formats are shown at the top of the file browser see Figure 2 Select the appropriate file format for your microarray data set and load the file Figure 2 Open File ata Format Optional _ Gene Expression Omnibus Series Matrix File GPCL Pre CLuster File Once loaded you will be asked whether yo2. false display previous clustering results input clustering output text file false normalize input by unit variance n1 to log2 n0 or into range 0 X nX false 1 perform median center normalization on all rows 2 columns 3 rows and columns 1 perform sum of squares 1 normalization on all rows 2 columns 3 rows and columns normalize input to log2 with unit variance false apply unit variance normalization to distance matrix false read in PCL formatted input file false read in Gene Expression Omnibus Series Matrix formatted input file false transform columns from input into Euclidean distance matrix false distance matrix metric 2 Pearson s 3 Uncentered Correlation Euclidean fill missing values 1 means of rows 2 medians of rows 3 means of columns 4 medians of columns means of rows do benchmarking F measure Precision Recall NMI corrected Rand Index data items must be labeled 1 2 3 total number of clusters false set number of Monte Carlo simulations for MST clustering 10 set x of SOM grid xy where y x Square root of input size 2 set maximum x y grid of SOM 30 set minimum x y grid of SOM 5 set SOM distance metric to Pearson Correlation Euclidean set SOM distance metric to Uncentered Correlation Euclidean set SOM topology to square circle set number of SOM iterations 1000 set SOM error surface exponent 3 set density equalizing cartogram resolution 64 disable Densi
3. by specifying an upper bound x iv Median Center Rows Arrays For microarray analysis centering each gene row and or array column by subtracting the median value of the row column eliminates amplitude shifts to cluster distinct expression patterns rather than overall transcript levels We recommend applying median centering to genes before co expression clustering clustering rows v Sum of Squares 1 Rows Arrays This normalization procedure smoothes the data set by forcing the sum of squares of all expression values to equal 1 for each row column in the data set The impact of this method on cluster identification can be significant and tends to result in the detection of larger clusters trailing off into genes with minimal differential expression after clustering the signal to noise ratio can be boosted by using the confidence filter see Output below As an example to run AutoSOME co expression analysis one may apply log2 scaling unit variance normalization and median centering of genes and or arrays To run AutoSOME transcriptome clustering fuzzy cluster networks see tutorial below it is generally recommended to apply unit variance normalization to your data set Other normalization settings may also be desirable Input can also be adjusted using Microsoft Excel or the Cluster software Eisen et al 1998 and then imported into AutoSOME Figure 6 Basic Fields Hide Inpu
4. data item identifier then click Submit If it is found your data item will be highlighted in yellow using the heat map display you might need to scroll down to find it Please make sure your data items are uniquely labeled for this feature to work properly Saving data 1 ii Images All images can be saved by selecting File gt Export gt save image from the menu Tabular Data To export tabular data from the selected cluster s go to File gt Export gt save tabular data This option will save all cluster contents along with cluster labels and cluster confidence for each data point You can choose between saving clusters with your original data pre normalized or with data after normalization has been applied current normalization The current normalization is dependent upon whether you have altered the heat map display using the Display Original Data feature in the image settings window Figure 13 The output format is identical to output file 3 see Files written to disk below Files written to disk AutoSOME writes the following files to your hard drive X No ensemble iterations e g E100 Y p value threshold e g Pval0 1 Z rows or columns depending upon the value of Cluster Analysis see pg 9 1 2 3 4 AutoSOME inputName EX PvalY Summary Z html summary table of file name parameters used and cluster output table with size of each cluster mean cluste
5. file format and save image i r To expedite this process for next time start with your saved network All visualization parameters will still be specified Then simply input your edge and node files and perform the layout Note that for the Node Color property it is important to assign colors again using the process given in step n 22 References 1 2 3 4 5 D haeseleer P How does gene expression clustering work Nature Biotechnology 2005 23 12 1499 1501 Eisen MB Spellman PT Brown PO Botstein D Cluster analysis and display of genome wide expression patterns Proc Natl Acad Sci USA 1998 95 25 14863 14868 Newman AM and Cooper JB AutoSOME a clustering method for identifying gene expression modules without prior knowledge of cluster number BMC Bioinformatics 2010 11 117 Saldanha A Java Treeview extensible visualization of microarray data Bioinformatics 2004 20 17 3246 3248 Shannon P Markiel A Ozier O Baliga NS Wang JT Ramage D Amin N Schwikowski B Ideker T Cytoscape A Software Environment for Integrated Models of Biomolecular Interaction Networks Genome Research 2003 13 2498 2504 23
6. AutoSOME User Manual February 16 2014 Written by Aaron M Newman Ph D Institute for Stem Cell Biology and Regenerative Medicine Stanford University Table of Contents O O O E ENEE 1 System Requirements eeaeee t ae EE EE A EEEE EE EE E d 1 Installing and Running the AutoSOME GUL ccccccsecsecsseceseceseceeeceeeeseneeeneesseeeseecsaecsaecseeeaeenaeees 1 Using Ato SOME sisene iaa e A e rar E AR EE A s i 2 Input Format A a a a aa aa a li A ist 2 Missme Values is tad 3 AUtOSOME GU cave nk ged en ete dened a A td dd Ra 3 Load and Filter Input cccccecccecssecseesecssecesecnsecesecnsecneeseeeeeeseneeeseeeseeeseeesaecsaecnseenseeneeees 3 Input Adjustment eai eaa e EE cacvadscudegvond A A A 6 Baste Fields Jeina a A A e a 7 Advanced Hilda ti aaa 8 Fuzzy Cluster Networks A T A 8 MIA O O EN 8 Al GOrithim Seton Gs it la tando tos biela edad dad lado bo dd e o Moo 8 Self Organizing Map SOM cc cesscsssesseessceescesecesecsecseecseeeseeeseeeeeeeenseeeseeaeenaeenaecsaeeaeeenes 8 Carte Srain eire ae e cee A AAA ias 9 Running AUtoSOME senise ime parse iaae ia aan aR ENa e Ea PEAD e SDE SEa 9 OUUP UE AAEE A TAE E A E 10 Output Elle E E E E aa 10 Csr T iia E RE EE EA E EN dada E nek 10 NN 11 A wtih Ae E RA eee MA a LRA eee 12 VIEWS Image AA Sec 12 View gt settings gt scale using entire data Set ceecesccssessecsseceseceecseeeseeeeeeeseeeeeesceeseeeseeeseenseesaeenees 13 CATCH gcse eater e sd desc enn shave sabe E
7. EE stu segeevte hace ENEN ENAA ENE rans sennsbceese 14 Saving data arna e Bhi Ah ee ee A eee eek DAL Ae ee ER ee 14 Piles written todisK ii Ai ii 14 Opening previous clustering results it a Ghd canta eae RA ea te ae 15 RES ii A A opvesceveyses stbedees A A EA 15 Enhanced GUl ys Basic GUP irnod a Rai 15 AutoSOME Command Line Version ecccesessssecesceseeseeeeceaecaeeecaecaeeeceesecaaeeneeaeceaeeaeeereaeeas 16 Creating Fuzzy Cluster Networks ccccccsccsssesseceteceseceseceeeseeeseneseseeeseecseeesaecsaecseeseeseeeseneenseeesees 18 IPLA ARAO a e E Pe a e a 23 Introduction AutoSOME is a powerful unsupervised computational method for identifying discrete and fuzzy clusters of diverse geometries from potentially large multi dimensional data sets without prior knowledge of cluster number or structure Newman and Cooper 2010 We have implemented the AutoSOME method as a Graphical User Interface GUI and command line tool to facilitate its use by the academic research community Below instructions are provided for using both versions of AutoSOME and a protocol is given for generating AutoSOME fuzzy cluster networks in Cytoscape Shannon et al 2005 Refer to the AutoSOME website for tutorials software and FAQ http jimcooperlab mcdb ucsb edu autosome System Requirements AutoSOME is coded in Java to promote platform independence To launch the GUI from the AutoSOME website http jimcooperlab mcdb ucsb edu autosome
8. Java Web Start technology needs to be installed on your computer along with Java Standard Edition 1 6 both of which are freely available at http java sun com The command line version of AutoSOME will run in JAVA 1 5 and perhaps earlier versions In general the more memory and processors available the better the performance of AutoSOME Without user intervention both the GUI and command line versions will greedily use all available CPUs For 32 bit Windows systems the maximum amount of memory that can be allocated to AutoSOME is about 1 6 GB while 64 bit operating systems with 64 bit Java can allocate up to 30 GB of memory Microarray data sets like the HG U133 Plus 2 0 Affymetrix chipset gt 54k probes with dozens of samples can be run with 1 6 GB RAM The Java Web Start launch button on the AutoSOME Web Portal homepage will automatically allocate up to 1 GB of RAM To run AutoSOME with more or less memory download the jar file from http jimcooperlab mcdb ucsb edu autosome downloads Installing and Running the AutoSOME GUI Download AutoSOME from the website and then unzip all contents of the downloaded archive into the same directory Using your system console navigate to the directory where you installed AutoSOME and type java Xmx1600m Xms1600m jar autosome_vXXXXXX jar XXXXXX the GUI version you downloaded each version is named by the date of compilation e g 120109 The Xmx Xms arguments allocate addi
9. This format contains all chipset expression data as well as user supplied annotation in a spreadsheet style AutoSOME can automatically extract expression information content and column names from a raw series matrix file This allows for rapid analysis of any GEO data set by simply downloading the archive containing the series matrix text file unzipping it and loading it into AutoSOME PCL Pre CLuster format used for the Cluster software Eisen et al 1998 PNAS 95 14863 For an example see http puma princeton edu help formats shtml pcl or the 2 Cluster TreeView User Manual The first two columns are reserved for gene annotation column 1 row identifiers column 2 gene names or annotation Column 3 is optional is called GWEIGHT and specifies how to weigh each gene when computing gene gene similarity This column is read by AutoSOME but is ignored since AutoSOME does not cluster genes using a similarity matrix Row 1 is mandatory and is used to provide column names including names of each array e g ID NAME GWEIGHT arrayl array2 Row 2 is optional is called EWEIGHT and specifies how each array is weighed when computing array array similarity In contrast to GWEIGHT AutoSOME will use EWEIGHT when constructing a distance matrix for transcriptome clustering fuzzy cluster networks option Missing Values There are two strategies implemented in AutoSOME for handling missing values blank space or NA
10. anges from 5 never co clustered to 0 5 always co clustered 6 AutoSOME inputName EX PvalY Nodes txt All clustered data points first column unique data point label second column cluster label third column original data point label 7 AutoSOME inputName EX PvalY Matrix txt pairwise affinity matrix of all data points compared to all data points essentially the same information content of output file 4 presented in a form suitable for a heat map display This file can be immediately read as input into the Cluster 3 0 software Eisen et al 1998 to hierarchically reorder the matrix Results can be visualized with Java TreeView Saldanha 2004 Since Fuzzy Cluster Networks are computed from a distance matrix of all vertical data vectors e g cell samples output files 1 3 and 6 contain data points represented as a distance matrix not the original input data Opening previous clustering results To browse previous clustering results select Open AutoSOME Results from the File menu Use the file browser to find the output text file either file 3 or 4 in files written to disk above and press Open The GUI will be immediately redirected to the output window and display the cluster tree All original data normalization settings are automatically restored for display in the output window Reset To reset all AutoSOME settings go to File gt Reset Enhanced GUI vs Basic GUI The enhanced and
11. basic GUIs offer identical functionality The look and feel of the enhanced GUI makes use of the Substance freeware package https substance dev java net while the basic GUI uses the default system look and feel 15 AutoSOME Command Line Version gt Usage java jar autosome vXXXXXX jar Input Options gt maximum JVM memory recommended e g java Xmx1600m Xms1600m jar autosome vXXXXXX jar Input Options where XXXXXX release date of AutoSOME e g 040110 To display all input parameters run AutoSOME with the parameter o letter o for options As indicated below the command line version of AutoSOME has many options not available in the GUI including the option to use alternative clustering algorithms i e K Means Hierarchical Clustering with four linkage types and alternative dimensional reduction techniques i e density equalized SOM normal SOM and Sammon s Mapping For alternative clustering methods the number of input clusters needs to be specified using the k parameter see below In addition all of these clustering methods including AutoSOME can be benchmarked with the b option as long as the data item labels in the input file correspond to numerical cluster labels starting with 1 1 2 No clusters 7 Examples using fictitious gene expression data set yeast txt 1 Perform co expression clustering using log2 scaling unit variance and median centering of rows normalization 100 e
12. ck and increase the NODE_LINE WIDTH to 2 Select Apply Using the VizMapper tab Next to Node Label click ID and select Column 3 All nodes should now be relabeled according to the original data labels Minimize the Node Label property by selecting the minus icon Under i ii iii iv Unused Properties find Edge Color and double click it Select Column 3 as a value Then select Continuous Mapper for Mapping Type Click on the black to white gradient next to Graphical View to launch a Gradient Editor There are two fixed triangles one on each end and two adjustable triangles Double click the two leftmost triangles and set their colors to pure red 255 0 0 Double click the two rightmost triangles and set their colors to pure blue 0 0 255 Drag the leftmost adjustable triangle all the way to the left and likewise drag the rightmost triangle to the right until it stops 19 l Find Edge Line Width under Unused Properties and double click it i ii iii iv Select Column 3 as a value Then select Continuous Mapper for Mapping Type Click on the graph next to the Graphical View property to launch the Continuous Editor Adjust the minimum and maximum values denoted by red squares double click on squares for precision otherwise slide squares up or down For example set m
13. e maximum and minimum va ues Maximum 14 9503 over the entire data set are also provided Minimum 2 485541 When finished clustering this table will dynamically update based on the contents of selected clusters in the cluster tree see Output below Input Adjustment AutoSOME implements several input normalization methods in addition to log2 scaling For an excellent overview of these techniques see the manual to the Cluster software available at http rana lbl gov manuals ClusterTreeView pdf Eisen et al 1998 Press the Show button to the right of the Input Adjustment label to access input scaling and normalization options see Figure 6 below All input adjustment operations are performed in the order listed in the GUI from top to bottom In brief i Log2 Scaling Logarithmic scaling is routinely used for microarray data sets to amplify small fold changes in gene expression and is completely reversible We recommend applying log2 scaling in cases where expression values span several orders of magnitude All other implemented input adjustment methods irreversibly change the input to make it more suitable for analysis ii Unit variance forces all columns to have zero mean and a standard deviation of one and is commonly used when there is no a priori reason to treat any column differently from any other iii Range 0 x Alternatively data in all columns can be normalized to share lowest and highest values 0 x
14. e sure you select a large Zoom Factor before saving Figure 13 Image Settings Dimensions 0 el Adjust Height B Zoom Factor O Heat Map Contrast c IE D Manually adjust range for contrast Minimum 000 Maximum 00 ea me Normalization Display Options _ Display Original Data 1 Log2 Scaling 2 Unit Variance 2 Range 0 x x 3 Median Center 4 Sum Squares 1 Both Pa A Normalization Display Options C Sort by Decreasing Variance Black Background _ Hide Heat Map Confidence Bar Hide Heat Map Row Labels Hide Heat Map Column Labels Hide Cluster Row Separators Oooo oH Hide Cluster Column Separators Hide Heat Map Color Bar View gt settings gt scale using entire dataset When selected go to View gt settings both the contrast in the heat map and y axis for signal plots are scaled using the maximum and minimum values of the data set When deselected maximum and minimum values are set based on the selected cluster s Deselecting is useful for amplifying cluster specific signals that are otherwise washed out in the context of the entire input data set 13 Search To find a particular data item in the cluster output go to Search gt Find This will invoke a search window Either find your data point from the Choose Identifier list of manually enter your
15. es for the x y length of the SOM node lattice ii Minimum Grid Length The minimum number of nodes for the x y length of the SOM node grid In general increase Maximum Grid Length to provide greater resolution for cluster separation in the SOM and to increase the number of possible clusters Increase Minimum Grid Length to force AutoSOME to use an SOM lattice with specific minimum dimensions Set both parameters to the same value to disable automatic adjustment of grid length by the algorithm The SOM grid length is set by default between 5 and 30 providing sufficient cluster resolving power for most clustering problems If a data set with at least 5 000 rows is imported e g whole genome microarray the maximum grid length will be automatically set to 20 for more efficient running time iii Training Iterations The number of iterations for training the SOM node lattice The value of this parameter dictates the number of iterations for each of two phases coarse grained followed by fine grained training This value can be toggled between 1000 iterations for precision used during benchmarking in Newman and Cooper 2010 500 iterations for normal and 250 iterations for speed iv Use Square Topology Select this checkbox to use a square shaped node lattice rather than a circular node lattice default Benchmarking results indicate both topologies yield comparable results See the Manuscript for more deta
16. he selected cluster s will be displayed as a list of cluster labels For additional viewing options see below Heat maps Several heat map visualizations are available Go to View in the menu bar and select green red rainbow gray scale or blue white from the heat map submenu Table 2 summarizes the key attributes of the available heat map color modes Table 2 Heat map High Value Low Value Middle Value s Mode R G B R G B green red Green 0 255 0 Red 255 0 0 Black rainbow Red 255 0 0 Blue 0 0 255 Blue Light Blue Green Yellow Red evenly spaced gray scale Black 0 0 0 White shades of Gray 255 255 255 blue white Blue 0 0 255 White shades of Blue 255 255 255 The mouse scrollbar is used to zoom in or out To fit the heat map to the screen go to view gt fit to screen or right click your mouse when hovering over the heat map display By default cluster confidence is shown as a vertical bar to the left of the heat map with blue high and red low confidence see rainbow colored heat map in Figure 11 below For fine grained control over heat map visualization parameters select view gt settings gt image settings see below 11 Signal plots As indicated in Figure 12 signal plots display all numerical vectors in the selected cluster s as a line graph across all column categories x axis Go to View in the menu bar and select rainb
17. ils Cartogram Resolution The density equalizing cartogram algorithm requires an input array with dimensions that are a power of two Although all benchmarking tests were conducted using an array size of 64X64 greater resolutions may yield more accurate density equalization especially when the SOM dimensions are large On the other hand going from 64X64 to 32X32 may increase running time dramatically up to 4X and is sufficient for accurate clustering results in many cases especially when the SOM dimensions are less than 30X30 Running AutoSOME After the input file and parameters are specified execute AutoSOME clustering by pressing the green RUN button in the AutoSOME Steps panel see Figure 1 Progress is shown in the lower left corner of the GUI and elapsed time is displayed see Figure 7 When finished the GUI will automatically be redirected to the output panel for browsing the cluster output The input and output panels can be toggled back and forth using the tabs or the INPUT and OUTPUT buttons in the control panel Figure 7 Run Progress 00 00 11 Computing Clusters Cancel Output Output Files The Output Files table located on the center left side of the GUI will show links to the AutoSOME output files after clustering is finished Figure 8 Simply select a file from the table and the file will be automatically displayed For details of the output files generated by A
18. inimum to 0 5 and maximum to 20 Then exit Continuous Editor m Find Bdge Opacity under Unused Properties and double click it i ii iii iv Select Column 3 as a value Then select Continuous Mapper for Mapping Type Click on the graph next to the Graphical View property to launch the Continuous Editor Adjust the minimum and maximum values denoted by red squares double click on squares for precision otherwise slide squares up or down For example set minimum to 0 5 and maximum to 60 Then exit soos Editor 20 n Finally find Node Color under Unused Properties and double click it i Select Column 2 as a value ii Then select Discrete Mapping for Mapping Type 111 Right click on Discrete Mapping and go to Generate Discrete Values gt Rainbow 1 All nodes are now colored according to cluster labels Adjust colors as necessary LAA o Goto Layout in the main menu and select Settings i Choose Force Directed Layout for Layout Algorithm ii Under Edge Weight Settings set The minimum edge weight to consider to 0 5 and set The maximum edge weight to consider to 0 5 Further select Column 3 from The edge attribute that contains the weights Layout Algorithm Force Directed Layout X Force Directed Layout Settings Standard setti
19. ngs Partition graph before layout Only layout selected nodes E Edge Weight Settings The edge attribute that contains the weights Column 3 How to interpret weight values Heuristic The minimum edge weight to consider in in la G The maximum edge weight to consider iii Press Execute Layout to run the layout algorithm iv Although the network topology is generally preserved different runs of the layout algorithm can yield slightly different results in terms of network rotation and local node placement To increase the repulsion between neighboring nodes for evenly spaced nodes within a cluster increase Default Node Mass under Algorithm settings 21 Another layout algorithm that can yield comparable results is the Edge weighted Spring Embedded algorithm Before executing the layout make sure the Edge Weight Settings are adjusted as in 11 above This layout algorithm can yield more evenly spaced nodes but is less stable than Force Directed Layout Run a few times v Notice that all edges are slightly curved To straighten edges save and reopen the Cytoscape file vi At this point make any desirable fine grained changes to the Edge Color Edge Line Width and Edge Opacity parameters to emphasize the fuzziness in the network p To export the final network go to File gt Export gt Network View as Graphics q Then select
20. nsemble runs p value threshold of 0 05 launch GUI to display results and write files to C output java Xmx1600m Xms1600m jar autosome_v120109 jar yeast txt N 1 e100 p 05 v DC output 2 Perform transcriptome clustering using unit variance Uncentered correlation 100 ensemble runs launch GUI to display results and write files to C output java Xmx1600m Xms1600m jar autosome_v120109 jar yeast txt nl Q3 e100 v DC output 3 Perform transcriptome clustering using unit variance Pearson s correlation to build the distance matrix K means with 5 clusters 20 ensemble iterations and launch GUI to display results java Xmx1600m Xms1600m jar autosome_v120109 jar yeast txt n1 Q2 K k5 e20 v 16 All Command Line Parameters Parameter t integer e integer p 0 1 D directory C V v2 n integer 1 2 3 u 1 2 3 N W W Q Q 2 3 h 1 2 3 4 b c integer g integer M integer m integer P P2 S integer x integer r power of 2 E S S integer k integer K A A method V Description Default set number of threads available CPUs set number of runs to merge into ensemble 10 set p value threshold for minimum spanning tree clustering 0 1 set output directory same directory as input file read in column headers from first row of input file auto detect otherwise launch cluster viewer
21. ow or red from the signal plot submenu i Scale bar By default a scale bar is shown on the y axis Deselect the scale bar checkbox in View gt signal plot to remove To increase or decrease the scale bar resolution use the up and down arrow keys in the number pad 8 and 2 respectively To change the decimal precision of each number use the left and right arrow keys in the number pad 4 and 6 11 Mean signal Select the mean signal checkbox in View gt signal plot Collapses signal plot into one line representing the mean of all vectors in the selected cluster s Figure 12 100 pa N AN View gt settings gt Image Settings The image settings window provides several options for customizing the heat map and signal plot display see Figure 13 below 1 Dimensions a Adjust Height Determines the vertical resolution of the image can be used to compress or expand image height The value of adjust height multiplied by the Zoom Factor see below gives the total number of vertical pixels for each heat map cell b Zoom Factor This slider directly controls the number of horizontal pixels in each heat map cell The overall image size is then adjusted by computing the dimensions width x height in pixels of each heat map cell Zoom Factor x Adjust Height Zoom Factor Use this option for saving high resolution images 11 Heat Map Contrast Adjusts the maximum and minimum values used for heat map dis
22. play Contrast C increases the original maximum value M and decreases the original minimum value m by C 1 M m 2 a Manually adjust range for contrast Select this option to manually input maximum and minimum values for contrast adjustment By default the maximum and minimum values of the entire input data set are used To see the maximum and minimum values for a particular cluster or set of clusters check the Input Data table on the left panel of the GUI 12 iii Normalization Tab By selecting Display Original Data the heat map will render the cluster s using your original input data pre normalization You can then select any of the data transformation options e g log2 and median centering to re normalize your data as desired This option is useful for displaying a more meaningful range of values in the heat map e g down to up regulated genes iv Display Options Tab Most of these options are self explanatory If you select Sort by Decreasing Variance the contents of every cluster will be sorted by decreasing order of row vector variance e g gene probe variance Although the heat map will update after changing most display settings you can force the image to refresh by pressing Update Click Save to launch a file browser for saving your heat map output All images are saved in the Portable Network Graphics PNG format Importantly if you want to output a high resolution image mak
23. r confidence and hyperlinks to data content of each cluster AutoSOME inputName EX PvalY Z html all clusters and data contents including individual cluster confidence for each data point AutoSOME inputName EX PvalY Z txt text file with same data as file 2 First row column labels if provided as input first column CLUST cluster label second column CONF cluster confidence third column NAME data point label all other columns data vectors AutoSOME inputName_ rows columns txt text file produced from clustering both rows and columns Is in same format as 3 with the exception of an additional row row 2 that stores cluster identifiers for the columns 14 If your data set was normalized using AutoSOME a code is stored using row 1 of output file 3 in order to properly reopen the file see Opening old clustering results below The code legend input data set has column labels n log2 scaling u unit variance sX normalized from 0 X m median centering of rows M median centering of columns q sum of squares normalization of rows Q sum of squares normalization of columns If columns are clustered see Cluster Analysis pg 9 three additional files are written to disk 5 AutoSOME inputName EX PvalY Edges txt fuzzy edges among data points first two columns denote connected nodes third column pairwise affinity or the fraction of times the two data points were co clustered pairwise affinity r
24. t Adjustment Hide Cluster Analysis Rows y 7 1 Log2 Scaling Running Mode Normal y goz Unit ae uN No Ensemble Runs 50 7 2 Range 0 x x 99 P value Threshold 01 7 3 Median Center Rows y No CPUs 12 sl 4 Sum Squares 1 Both y Advanced Fields Hide Algorithm Settings Hide Fuzzy Cluster Networks SOM Column Clustering Maximum Grid Length 30 PE ietie Minimum Grid Length 5 Euclidean Distance Training Iterations 500 y nit Variance Normalize C Use Square Topology Circular Default Memory Cartogram C Write Ensemble Runs to Disk Resolution 32x32 y Basic Fields Press the Show button to expand the Basic Fields section see Figure 6 i Cluster Analysis Switch among clustering rows e g genes columns e g experiments or both rows and columns in a single step 11 Running Mode This parameter alters advanced AutoSOME algorithm settings to switch among Precision Normal and Speed modes of operation Precision takes longer but provides greater training of the SOM node lattice 2X1000 iterations and greater resolution for density equalization of the SOM error surface 64X64 For enhanced performance and especially for first pass exploratory cluster analysis choose Speed for less SOM iterations 2X250 and less resolution for density equalization 16X16 Normal provides a compromise between the other two settings 2X500 SOM iterations and 32X32 cartogram resol
25. tional memory to the 1 Java Virtual Machine necessary for running large data sets Allocate more or less memory as needed for your input data set parameters and machine architecture Also AutoSOME has not been internationalized If it is behaving oddly try adding the following JVM argument Duser language en For instructions on how to use the AutoSOME command line interface see AutoSOME Command Line Version below Using AutoSOME Input Format AutoSOME accepts three kinds of numerical input files Standard format The input file is a table of numerical values with optional column labels row 1 and mandatory row labels column 1 If column labels are specified column 1 also needs a label in row 1 All entries must be tab comma or space delimited Input data can be easily formatted using Microsoft Excel See Table 1 below for an example Table 1 Probe hESC 1 hESC 2 hESC 3 hESC 4 iPSC 1 iPSC 2 iPSC 3 212853 at 8 221449 8 297634 7 694108 8 215596 10 1284 10 25488 10 21546 212854_x_at 8 523748 8 706556 8 044123 8 992252 8 87927 8 974617 9 083473 212855 at 10 64296 10 4093 10 60148 11 03563 12 08599 11 91321 12 04592 212856 at 8 986721 9 218977 9 109346 8 425392 8 431269 8 418733 8 450901 212857_x_at 6 195244 5 929077 4 432911 6 764105 4 344787 4 786857 4 953329 Microarray formats 1 Gene Expression Omnibus GEO Series Matrix
26. tions of Euclidean Pearson s and Uncentered correlation distance metrics In brief Euclidean distance is sensitive to amplitude shifts while Pearson s correlation is not Experiment with these metrics to get a feel for how they work If smoothing out the distance matrix is desired Unit Variance Normalize can be selected to normalize the distance matrix columns to unit variance When finished clustering AutoSOME will automatically generate additional output files for building fuzzy cluster networks in Cytoscape see Fuzzy Cluster Networks below Memory Write Intermediate Runs to Disk If AutoSOME crashes or seems to hang for a long period of time then Write Ensemble Runs to Disk should be selected see Figure 6 as an out of memory error is likely This may be necessary on systems with insufficient RAM for large data sets and many ensemble runs e g 45k probes 100 samples and 2000 ensemble runs Overall running time will be slower due to reading and writing to disk Files will be written to a temporary folder in the output directory Also try reducing the number of CPUs to free up memory Algorithm Settings Press the Show button to expand the Algorithm Settings section see Figure 6 The GUI permits modification of the most critical algorithm parameters For access to all algorithm parameters use the command line version of AutoSOME Self Organizing Map SOM i Maximum Grid Length The maximum number of nod
27. ty Equalizing Cartogram false invoke Sammon Mapping instead of SOM false set number of Sammon Mapping iterations 100 specify number of clusters in data set false invoke K Means Clustering requires option k false invoke Agglomerative Clustering requires option k false 1 Single 2 Complete 3 Average 4 Ward s 4 print verbose output false 17 Creating Fuzzy Cluster Networks Fuzzy cluster networks highlight the fuzzy relationships among clustered data points using an intuitive two dimensional network display A powerful application of this approach is the visualization of differences between cell lines on the basis of differential gene expression To display fuzzy cluster networks the network visualization tool Cytoscape needs to be installed on your computer Shannon et al 2005 Cytoscape is freely available from http www cytoscape org 1 2 3 4 5 Import your input file into AutoSOME set fields and adjust input Select Fuzzy Cluster Networks and choose distance metric Please remember that only vertical data vectors are clustered e g cell samples or time series of a microarray data set Run AutoSOME Launch Cytoscape In Cytoscape All screenshots below taken from Cytoscape 2 6 0 a Go to File gt Import gt Network from Table Text MS Excel b d Select Import and then Close A raw network will appear as a grid Go to Select File s and locate AutoSOME o
28. u want to filter your input data set Figure 3 Figure 3 Data Summary Input Successfully Loaded Rows 2093 Columns 62 Maximum 5 83 Minimum 5 57 Do you want to filter your data before clustering If you choose yes you will be redirected to a Filter Data window Figure 4 below This window provides two simple data filtration options i Remove rows with fold change less than X The fold change is calculated by the difference between the minimum and maximum values for each row All rows with fold change lt X will be removed Importantly if your data are already log2 scaled make sure you select the corresponding checkbox 11 Remove rows with mean value below Y The mean u of each row is calculated and the row is removed if u lt Y After specifying filtration options press Apply to preview the filtered data and if you approve press Accept Figure 4 Filter Data Original Data Rows 2093 Columns 62 Maximum 5 83 Minimum ta are already Log2 scaled Y Remove rows with fold change less than 4 C Remove all rows with mean value below Filtered Data after removal Rows 2093 Maximum 5 83 Minimum 5 57 Figure 5 After your input file has loaded the Input Data number of rows and columns will be displayed in the Input Data panel on Property Value p ay pu p the left side of the GUI see Figure 5 Rows 54675 Th d ban 1 Columns
29. ution In our experiments Normal works quite well and in fact often yields comparable results to Precision Choose Speed for a first pass exploratory analysis iii No Ensemble Runs The default of 50 ensemble iterations should be sufficient to begin investigating the cluster structure of most data sets Although in practice AutoSOME clustering results can be quite stable with 50 ensemble iterations and even as little as 20 for final clustering results it is recommended to increase this number to at least 100 iv P value Threshold AutoSOME has been extensively benchmarked on a highly diverse array of clustering problems using a P value cutoff of 0 1 Reduce the p value threshold to identify tighter clusters v No CPUs To liberate processor resources decrease No CPUs AutoSOME running time will decrease approximately linearly with respect to increasing number of dedicated CPUs Advanced Fields Fuzzy Cluster Networks When clustering columns i e Cluster Analysis in Basic Fields is set to columns or both the Fuzzy Cluster Networks window will be automatically enabled see Figure 6 In this window you can pick one of three distance metrics for calculating the input distance matrix Euclidean Pearson s and Uncentered correlation Euclidean distance is selected by default due to good results obtained from empirical testing See the review by D haeseleer 2005 for descrip
30. utoSOME see files written to disk below Figure 8 Output Files File Type Summary html Clusters html Clusters text Cluster tree As shown in Figure 9 all clusters are listed using a dynamic tree structure and are ordered from top to bottom by decreasing size Figure 9 Input Output de clusters de cluster 1 17 cluster 2 10 Jo cluster 3 6 mn cluster 4 2 Once a cluster node is selected using your mouse the cluster tree can be rapidly traversed by pressing the f and keys To select more than one cluster hold down the Shift or Control key and select using your mouse Clusters can be manually modified using the control panel below the cluster tree Figure 10 Use Split to make a new cluster from a selected group of data points first expand a cluster node by double clicking then select data points with mouse Press Delete key to erase entire clusters or specific data points Press Merge to combine the contents of all selected clusters into a single cluster Press Reset to return to the original clustering output To filter clustering results by cluster confidence metric ranging from 1 100 where 100 data point always in cluster x type in confidence threshold lt 100 in the Confidence text box and press Update Figure 10 Split Merge Reset Confidence 0 Update 10 By default the contents of t
31. utput file 4 containing all edges see Files written to disk in Output above Set Source Interaction to Column 1 and Target Interaction to Column 2 Finally click on Column 3 in the data Preview window to activate it it will turn blue AutoSOME_golub_autFormat_E250_Pval0 1_Edges txt Y Column 1 Y Column2 V Column 3 _ B cell 0 B cell 1 _0 454545455_ B cell 0 B cell 2 0 363636364 B cell 0 B cell 3 _0 05785124 B cell 0 B cell 4 0 421487603 B cell 0 B cell 5 0 132231405 B cell 0 B cell 6 0 466942149 B cell 0 B cell 7 0 409090909 B cell 0 B cell 8 0 462809917 B cell 0 B cell 9 136363636 18 4 0 0 la n ii Pa L ve E e 6 4 N o o 0 o e 4 0009 10 000 0 0 8 K h Go to File gt Import gt Attribute from Table Text MS Excel Go to Select File s and locate AutoSOME output file 5 containing all nodes and attributes see Files written to disk in Output above Select Import Change global properties of the network i ii iii iv vi In the Control Panel select the VizMapper tab Click in the Defaults window shows a source and target pair with a blue background A new window will appear Select the Global tab in the bottom right Change the background color to white Go back to the Node tab and change the NODE BORDER COLOR property to bla
Download Pdf Manuals
Related Search