Mercurial > repos > yhoogstrate > edger_with_design_matrix
view edgeR_Differential_Gene_Expression.xml @ 50:228867d5283b draft
Uploaded
| author | yhoogstrate |
|---|---|
| date | Tue, 03 Jun 2014 05:30:13 -0400 |
| parents | f710e5ed7cea |
| children | b89788eead3c |
line wrap: on
line source
<?xml version="1.0" encoding="UTF-8"?> <tool id="edger_dge" name="edgeR: Differential Gene(Expression) Analysis"> <description>RNA-Seq gene expression analysis using edgeR (R package)</description> <requirements> <requirement type="package" version="3.0.3">R</requirement> <requirement type="package" version="latest">package_biocLite_edgeR_limma</requirement> </requirements> <command> <!-- The following script is written in the "Cheetah" language: http://www.cheetahtemplate.org/docs/users_guide_html_multipage/contents.html --> R --vanilla --slave -f $R_script '--args $expression_matrix $design_matrix $contrast $fdr $output_count_edgeR $output_cpm /dev/null <!-- Calculation of FPKM/RPKM should come here --> #if $output_raw_counts: $output_raw_counts #else: /dev/null #end if #if $output_MDSplot: $output_MDSplot #else: /dev/null #end if #if $output_BCVplot: $output_BCVplot #else: /dev/null #end if #if $output_MAplot: $output_MAplot #else: /dev/null #end if #if $output_PValue_distribution_plot: $output_PValue_distribution_plot #else: /dev/null #end if #if $output_hierarchical_clustering_plot: $output_hierarchical_clustering_plot #else: /dev/null #end if #if $output_heatmap_plot: $output_heatmap_plot #else: /dev/null #end if #if $output_RData_obj: $output_RData_obj #else: /dev/null #end if ' #if $output_R: > $output_R #else: > /dev/null #end if 2> stderr.txt ; grep -v 'Calculating library sizes from column' stderr.txt 1>&2 </command> <inputs> <param name="expression_matrix" type="data" format="tabular" label="Expression (read count) matrix" /> <param name="design_matrix" type="data" format="tabular" label="Design matrix" hepl="Ensure your samplenames are identical to those in the expression matrix. Preferentially, create the contrast matrix using 'edgeR: Design- from Expression matrix'." /> <param name="contrast" type="text" label="Contrast (biological question)" help="e.g. 'tumor-normal' or '(G1+G2)/2-G3' using the factors chosen in the design matrix. Read the 'makeContrasts' manual from Limma package for more info: http://www.bioconductor.org/packages/release/bioc/html/limma.html and http://www.bioconductor.org/packages/release/bioc/vignettes/limma/inst/doc/usersguide.pdf." /> <param name="fdr" type="float" min="0" max="1" value="0.05" label="False Discovery Rate (FDR)" /> <param name="outputs" type="select" label="Optional desired outputs" multiple="true" display="checkboxes"> <option value="make_output_raw_counts">Raw counts table</option> <option value="make_output_MDSplot">MDS-plot</option> <option value="make_output_BCVplot">BCV-plot</option> <option value="make_output_MAplot">MA-plot</option> <option value="make_output_PValue_distribution_plot">P-Value distribution plot</option> <option value="make_output_hierarchical_clustering_plot">Hierarchical custering</option> <option value="make_output_heatmap_plot">Heatmap</option> <option value="make_output_R_stdout">R stdout</option> <option value="make_output_RData_obj">R Data object</option> </param> </inputs> <configfiles> <configfile name="R_script"> library(limma,quietly=TRUE) ## enable quietly to avoid unnecessaity stderr dumping library(edgeR,quietly=TRUE) ## enable quietly to avoid unnecessaity stderr dumping library(splines,quietly=TRUE) ## enable quietly to avoid unnecessaity stderr dumping ## Fetch commandline arguments args <- commandArgs(trailingOnly = TRUE) expression_matrix_file = args[1] design_matrix_file = args[2] contrast = args[3] fdr = args[4] output_count_edgeR = args[5] output_cpm = args[6] output_xpkm = args[7] ##FPKM file - yet to be implemented output_raw_counts = args[8] output_MDSplot = args[9] output_BCVplot = args[10] output_MAplot = args[11] output_PValue_distribution_plot = args[12] output_hierarchical_clustering_plot = args[13] output_heatmap_plot = args[14] output_RData_obj = args[15] library(edgeR) ##raw_data <- read.delim(designmatrix,header=T,stringsAsFactors=T) ## Obtain read-counts expression_matrix <- read.delim(expression_matrix_file,header=T,stringsAsFactors=F,row.names=1,check.names=FALSE,na.strings=c("")) design_matrix <- read.delim(design_matrix_file,header=T,stringsAsFactors=F,row.names=1,check.names=FALSE,na.strings=c("")) colnames(design_matrix) <- make.names(colnames(design_matrix)) for(i in 1:ncol(design_matrix)) { old = design_matrix[,i] design_matrix[,i] = make.names(design_matrix[,i]) if(paste(design_matrix[,i],collapse="\t") != paste(old,collapse="\t")) { print("Renaming of factors:") print(old) print("To:") print(design_matrix[,i]) } ## The following line seems to malfunction the script: ##design_matrix[,i] <- as.factor(design_matrix[,i]) } ## 1) In the expression matrix, you only want to have the samples described in the design matrix columns <- match(rownames(design_matrix),colnames(expression_matrix)) columns <- columns[!is.na(columns)] read_counts <- expression_matrix[,columns] ## 2) In the design matrix, you only want to have samples of which you really have the counts columns <- match(colnames(expression_matrix),rownames(design_matrix)) columns <- columns[!is.na(columns)] design_matrix <- design_matrix[columns,,drop=FALSE] ## Filter for HTSeq predifined counts: exclude_HTSeq <- c("no_feature","ambiguous","too_low_aQual","not_aligned","alignment_not_unique") exclude_DEXSeq <- c("_ambiguous","_empty","_lowaqual","_notaligned") exclude <- match(c(exclude_HTSeq, exclude_DEXSeq),rownames(read_counts)) exclude <- exclude[is.na(exclude)==0] if(length(exclude) != 0) { read_counts <- read_counts[-exclude,] } empty_samples <- apply(read_counts,2,function(x) sum(x) == 0) if(sum(empty_samples) > 0) { write(paste("There are ",sum(empty_samples)," empty samples found:",sep=""),stderr()) write(colnames(read_counts)[empty_samples],stderr()) } else { dge <- DGEList(counts=read_counts,genes=rownames(read_counts)) formula <- paste(c("~0",make.names(colnames(design_matrix))),collapse = " + ") design_matrix_tmp <- design_matrix colnames(design_matrix_tmp) <- make.names(colnames(design_matrix_tmp)) design <- model.matrix(as.formula(formula),design_matrix_tmp) rm(design_matrix_tmp) # Filter prefixes prefixes = colnames(design_matrix)[attr(design,"assign")] avoid = nchar(prefixes) == nchar(colnames(design)) replacements = substr(colnames(design),nchar(prefixes)+1,nchar(colnames(design))) replacements[avoid] = colnames(design)[avoid] colnames(design) = replacements # Do normalization write("Calculating normalization factors...",stdout()) dge <- calcNormFactors(dge) write("Estimating common dispersion...",stdout()) dge <- estimateGLMCommonDisp(dge,design) write("Estimating trended dispersion...",stdout()) dge <- estimateGLMTrendedDisp(dge,design) write("Estimating tagwise dispersion...",stdout()) dge <- estimateGLMTagwiseDisp(dge,design) if(output_MDSplot != "/dev/null") { write("Creating MDS plot",stdout()) ##points <- plotMDS(dge,method="bcv",labels=rep("",nrow(dge\$samples)))# Get coordinates of unflexible plot points <- plotMDS.DGEList(dge,labels=rep("",nrow(dge\$samples)))# Get coordinates of unflexible plot dev.off()# Kill it pdf(output_MDSplot) diff_x <- abs(max(points\$x)-min(points\$x)) diff_y <-(max(points\$y)-min(points\$y)) plot(c(min(points\$x),max(points\$x) + 0.45 * diff_x), c(min(points\$y) - 0.05 * diff_y,max(points\$y) + 0.05 * diff_y), main="edgeR MDS Plot",type="n", xlab="BCV distance 1", ylab="BCV distance 2") points(points\$x,points\$y,pch=20) text(points\$x, points\$y,rownames(dge\$samples),cex=0.7,col="gray",pos=4) rm(diff_x,diff_y) dev.off() } if(output_BCVplot != "/dev/null") { write("Creating Biological coefficient of variation plot",stdout()) pdf(output_BCVplot) plotBCV(dge, cex=0.4, main="edgeR: Biological coefficient of variation (BCV) vs abundance") dev.off() } write("Fitting GLM...",stdout()) fit <- glmFit(dge,design) write(paste("Performing likelihood ratio test: ",contrast,sep=""),stdout()) cont <- c(contrast) cont <- makeContrasts(contrasts=cont, levels=design) lrt <- glmLRT(fit, contrast=cont[,1]) write(paste("Exporting to file: ",output_count_edgeR,sep=""),stdout()) write.table(file=output_count_edgeR,topTags(lrt,n=nrow(read_counts))\$table,sep="\t",row.names=TRUE,col.names=NA) write.table(file=output_cpm,cpm(dge,normalized.lib.sizes=TRUE),sep="\t",row.names=TRUE,col.names=NA) ## todo EXPORT FPKM write.table(file=output_raw_counts,dge\$counts,sep="\t",row.names=TRUE,col.names=NA) if(output_MAplot != "/dev/null" || output_PValue_distribution_plot != "/dev/null") { etable <- topTags(lrt, n=nrow(dge))\$table etable <- etable[order(etable\$FDR), ] if(output_MAplot != "/dev/null") { write("Creating MA plot...",stdout()) pdf(output_MAplot) with(etable, plot(logCPM, logFC, pch=20, main="edgeR: Fold change vs abundance")) with(subset(etable, FDR < fdr), points(logCPM, logFC, pch=20, col="red")) abline(h=c(-1,1), col="blue") dev.off() } if(output_PValue_distribution_plot != "/dev/null") { write("Creating P-value distribution plot...",stdout()) pdf(output_PValue_distribution_plot) expressed_genes <- subset(etable, PValue < 0.99) h <- hist(expressed_genes\$PValue,breaks=nrow(expressed_genes)/15,main="Binned P-Values (< 0.99)") center <- sum(h\$counts) / length(h\$counts) lines(c(0,1),c(center,center),lty=2,col="red",lwd=2) k <- ksmooth(h\$mid, h\$counts) lines(k\$x,k\$y,col="red",lwd=2) rmsd <- (h\$counts) - center rmsd <- rmsd^2 rmsd <- sum(rmsd) rmsd <- sqrt(rmsd) text(0,max(h\$counts),paste("e=",round(rmsd,2),sep=""),pos=4,col="blue") ## change e into epsilon somehow dev.off() } } if(output_heatmap_plot != "/dev/null") { pdf(output_heatmap_plot,width=10.5) etable2 <- topTags(lrt, n=100)\$table order <- rownames(etable2) cpm_sub <- cpm(dge,normalized.lib.sizes=TRUE,log=TRUE)[as.numeric(order),] heatmap(t(cpm_sub)) dev.off() } ##output_hierarchical_clustering_plot = args[13] if(output_RData_obj != "/dev/null") { save.image(output_RData_obj) } write("Done!",stdout()) } </configfile> </configfiles> <outputs> <data format="tabular" name="output_count_edgeR" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - differtially expressed genes" /> <data format="tabular" name="output_cpm" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - CPM" /> <data format="tabular" name="output_raw_counts" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - raw counts"> <filter>("make_output_raw_counts" in outputs)</filter> </data> <data format="pdf" name="output_MDSplot" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - MDS-plot"> <filter>("make_output_MDSplot" in outputs)</filter> </data> <data format="pdf" name="output_BCVplot" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - BCV-plot"> <filter>("make_output_BCVplot" in outputs)</filter> </data> <data format="pdf" name="output_MAplot" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - MA-plot"> <filter>("make_output_MAplot" in outputs)</filter> </data> <data format="pdf" name="output_PValue_distribution_plot" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - P-Value distribution"> <filter>("make_output_PValue_distribution_plot" in outputs)</filter> </data> <data format="pdf" name="output_hierarchical_clustering_plot" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - Hierarchical custering"> <filter>("make_output_hierarchical_clustering_plot" in outputs)</filter> </data> <data format="pdf" name="output_heatmap_plot" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - Heatmap"> <filter>("make_output_heatmap_plot" in outputs)</filter> </data> <data format="RData" name="output_RData_obj" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - R data object"> <filter>("make_output_RData_obj" in outputs)</filter> </data> <data format="txt" name="output_R" label="edgeR DGE on ${design_matrix.hid}: ${design_matrix.name} - R output (debug)" > <filter>("make_output_R_stdout" in outputs)</filter> </data> </outputs> <help> edgeR: Differential Gene(Expression) Analysis ############################################# Overview -------- Differential expression analysis of RNA-seq and digital gene expression profiles with biological replication. Uses empirical Bayes estimation and exact tests based on the negative binomial distribution. Also useful for differential signal analysis with other types of genome-scale count data [1]. For every experiment, the algorithm requires a design matrix. This matrix describes which samples belong to which groups. More details on this are given in the edgeR manual: http://www.bioconductor.org/packages/2.12/bioc/vignettes/edgeR/inst/doc/edgeRUsersGuide.pdf and the limma manual. Because the creation of a design matrix can be complex and time consuming, especially if no GUI is used, this package comes with an alternative tool which can help you with it. This tool is called *edgeR Design Matrix Creator*. If the appropriate design matrix (with corresponding links to the files) is given, the correct contrast ( http://en.wikipedia.org/wiki/Contrast_(statistics) ) has to be given. If you have for example two groups, with an equal weight, you would like to compare either "g1~g2" or "normal~cancer". The test function makes use of a MCF7 dataset used in a study that indicates that a higher sequencing depth is not neccesairily more important than a higher amount of replaciates[2]. Input ----- Expression matrix ^^^^^^^^^^^^^^^^^ :: Geneid "\t" Sample-1 "\t" Sample-2 "\t" Sample-3 "\t" Sample-4 [...] "\n" SMURF "\t" 123 "\t" 21 "\t" 34545 "\t" 98 ... "\n" BRCA1 "\t" 435 "\t" 6655 "\t" 45 "\t" 55 ... "\n" LINK33 "\t" 4 "\t" 645 "\t" 345 "\t" 1 ... "\n" SNORD78 "\t" 498 "\t" 65 "\t" 98 "\t" 27 ... "\n" [...] *Note: Make sure the number of columns in the header is identical to the number of columns in the body.* Design matrix ^^^^^^^^^^^^^ :: Sample "\t" Condition "\t" Ethnicity "\t" Patient "\t" Batch "\n" Sample-1 "\t" Tumor "\t" European "\t" 1 "\t" 1 "\n" Sample-2 "\t" Normal "\t" European "\t" 1 "\t" 1 "\n" Sample-3 "\t" Tumor "\t" European "\t" 2 "\t" 1 "\n" Sample-4 "\t" Normal "\t" European "\t" 2 "\t" 1 "\n" Sample-5 "\t" Tumor "\t" African "\t" 3 "\t" 1 "\n" Sample-6 "\t" Normal "\t" African "\t" 3 "\t" 1 "\n" Sample-7 "\t" Tumor "\t" African "\t" 4 "\t" 2 "\n" Sample-8 "\t" Normal "\t" African "\t" 4 "\t" 2 "\n" Sample-9 "\t" Tumor "\t" Asian "\t" 5 "\t" 2 "\n" Sample-10 "\t" Normal "\t" Asian "\t" 5 "\t" 2 "\n" Sample-11 "\t" Tumor "\t" Asian "\t" 6 "\t" 2 "\n" Sample-12 "\t" Normal "\t" Asian "\t" 6 "\t" 2 "\n" *Note: Avoid factor names that are (1) numerical, (2) contain mathematical symbols and preferebly only use letters.* Contrast ^^^^^^^^ The contrast represents the biological question. There can be many questions asked, e.g.: - Tumor-Normal - African-European - 0.5*(Control+Placebo) / Treated Installation ------------ This tool requires no specific configurations. The following dependencies are installed automatically: - R - Bioconductor - limma - edgeR License ------- - R - GPL-2 & GPL-3 - limma - GPL (>=2) - edgeR - GPL (>=2) References ---------- EdgeR ^^^^^ **[1] edgeR: a Bioconductor package for differential expression analysis of digital gene expression data.** *Mark D. Robinson, Davis J. McCarthy and Gordon K. Smyth* - Bioinformatics (2010) 26 (1): 139-140. - http://www.bioconductor.org/packages/2.12/bioc/html/edgeR.html - http://dx.doi.org/10.1093/bioinformatics/btp616 - http://www.bioconductor.org/packages/release/bioc/html/edgeR.html Test-data (MCF7) ^^^^^^^^^^^^^^^^ **[2] RNA-seq differential expression studies: more sequence or more replication?** *Yuwen Liu, Jie Zhou and Kevin P. White* - Bioinformatics (2014) 30 (3): 301-304. - http://www.ncbi.nlm.nih.gov/pubmed/24319002 - http://dx.doi.org/10.1093/bioinformatics/btt688 Contact ------- The tool wrapper has been written by Youri Hoogstrate from the Erasmus Medical Center (Rotterdam, Netherlands) on behalf of the Translational Research IT (TraIT) project: http://www.ctmm.nl/en/programmas/infrastructuren/traitprojecttranslationeleresearch I would like to thank Hina Riaz - Naz Khan for her helpful contribution. More tools by the Translational Research IT (TraIT) project can be found in the following repository: http://testtoolshed.g2.bx.psu.edu/ </help> </tool>