Mercurial > repos > vandelj > giant_hierarchical_clustering
view src/LIMMAscriptV4.R @ 0:14045c80a222 draft
"planemo upload for repository https://github.com/juliechevalier/GIANT/tree/master commit cb276a594444c8f32e9819fefde3a21f121d35df"
author | vandelj |
---|---|
date | Fri, 26 Jun 2020 09:38:23 -0400 |
parents | |
children |
line wrap: on
line source
# A command-line interface for LIMMA to use with Galaxy # written by Jimmy Vandel # one of these arguments is required: # # initial.options <- commandArgs(trailingOnly = FALSE) file.arg.name <- "--file=" script.name <- sub(file.arg.name, "", initial.options[grep(file.arg.name, initial.options)]) script.basename <- dirname(script.name) source(file.path(script.basename, "utils.R")) source(file.path(script.basename, "getopt.R")) #addComment("Welcome R!") # setup R error handling to go to stderr options( show.error.messages=F, error = function () { cat(geterrmessage(), file=stderr() ); q( "no", 1, F ) } ) # we need that to not crash galaxy with an UTF8 error on German LC settings. loc <- Sys.setlocale("LC_MESSAGES", "en_US.UTF-8") loc <- Sys.setlocale("LC_NUMERIC", "C") #get starting time start.time <- Sys.time() options(stringAsfactors = FALSE, useFancyQuotes = FALSE) args <- commandArgs() # get options, using the spec as defined by the enclosed list. # we read the options from the default: commandArgs(TRUE). spec <- matrix(c( "dataFile", "i", 1, "character", "factorInfo","a", 1, "character", "blockingInfo","b", 1, "character", "dicoRenaming","g",1,"character", "blockingPolicy","u", 1, "character", "fdrThreshold","t", 1, "double", "thresholdFC","d", 1, "double", "format", "f", 1, "character", "histo","h", 1, "character", "volcano","v", 1, "character", "factorsContrast","r", 1, "character", "contrastNames","p", 1, "character", "firstGroupContrast","m", 1, "character", "secondGroupContrast","n", 1, "character", "controlGroups","c", 1, "character", "fratioFile","s",1,"character", "organismID","x",1,"character", "rowNameType","y",1,"character", "quiet", "q", 0, "logical", "log", "l", 1, "character", "outputFile" , "o", 1, "character", "outputDfFile" , "z", 1, "character"), byrow=TRUE, ncol=4) opt <- getopt(spec) # enforce the following required arguments if (is.null(opt$log)) { addComment("[ERROR]'log file' is required\n") q( "no", 1, F ) } addComment("[INFO]Start of R script",T,opt$log,display=FALSE) if (is.null(opt$dataFile)) { addComment("[ERROR]'dataFile' is required",T,opt$log) q( "no", 1, F ) } if (!is.null(opt$blockingInfo) && is.null(opt$blockingPolicy) ) { addComment("[ERROR]blocking policy is missing",T,opt$log) q( "no", 1, F ) } if (is.null(opt$dicoRenaming)) { addComment("[ERROR]renaming dictionnary is missing",T,opt$log) q( "no", 1, F ) } if (is.null(opt$factorsContrast)) { addComment("[ERROR]factor informations are missing",T,opt$log) q( "no", 1, F ) } if (length(opt$firstGroupContrast)!=length(opt$secondGroupContrast)) { addComment("[ERROR]some contrast groups seems to be empty",T,opt$log) q( "no", 1, F ) } if (is.null(opt$factorInfo)) { addComment("[ERROR]factors info is missing",T,opt$log) q( "no", 1, F ) } if (is.null(opt$format)) { addComment("[ERROR]'output format' is required",T,opt$log) q( "no", 1, F ) } if (is.null(opt$fdrThreshold)) { addComment("[ERROR]'FDR threshold' is required",T,opt$log) q( "no", 1, F ) } if (is.null(opt$outputFile) || is.null(opt$outputDfFile)){ addComment("[ERROR]'output files' are required",T,opt$log) q( "no", 1, F ) } if (is.null(opt$thresholdFC)){ addComment("[ERROR]'FC threshold' is required",T,opt$log) q( "no", 1, F ) } if (is.null(opt$fratioFile)) { addComment("[ERROR]F-ratio parameter is missing",T,opt$log) q( "no", 1, F ) } #demande si le script sera bavard verbose <- if (is.null(opt$quiet)) { TRUE }else{ FALSE } #paramètres internes #pour savoir si on remplace les FC calculés par LIMMA par un calcul du LS-MEAN (ie moyenne de moyennes de chaque groupe dans chaque terme du contraste plutôt qu'une moyenne globale dans chaque terme) useLSmean=FALSE addComment("[INFO]Parameters checked!",T,opt$log,display=FALSE) addComment(c("[INFO]Working directory: ",getwd()),TRUE,opt$log,display=FALSE) addComment(c("[INFO]Command line: ",args),TRUE,opt$log,display=FALSE) #directory for plots dir.create(file.path(getwd(), "plotDir")) dir.create(file.path(getwd(), "plotLyDir")) #charge des packages silencieusement suppressPackageStartupMessages({ library("methods") library("limma") library("biomaRt") library("ggplot2") library("plotly") library("stringr") library("RColorBrewer") }) #chargement du fichier dictionnaire de renommage renamingDico=read.csv(file=file.path(getwd(), opt$dicoRenaming),header=F,sep="\t",colClasses="character") rownames(renamingDico)=renamingDico[,2] #chargement des fichiers en entrée expDataMatrix=read.csv(file=file.path(getwd(), opt$dataFile),header=F,sep="\t",colClasses="character") #remove first row to convert it as colnames (to avoid X before colnames with header=T) colNamesData=expDataMatrix[1,-1] expDataMatrix=expDataMatrix[-1,] #remove first colum to convert it as rownames rowNamesData=expDataMatrix[,1] expDataMatrix=expDataMatrix[,-1] if(is.data.frame(expDataMatrix)){ expDataMatrix=data.matrix(expDataMatrix) }else{ expDataMatrix=data.matrix(as.numeric(expDataMatrix)) } dimnames(expDataMatrix)=list(rowNamesData,colNamesData) #test the number of rows that are constant in dataMatrix nbConstantRows=length(which(unlist(apply(expDataMatrix,1,var))==0)) if(nbConstantRows>0){ addComment(c("[WARNING]",nbConstantRows,"rows are constant across conditions in input data file"),T,opt$log,display=FALSE) } #test if all condition names are present in dico if(!all(colnames(expDataMatrix) %in% rownames(renamingDico))){ addComment("[ERROR]Missing condition names in renaming dictionary",T,opt$log) q( "no", 1, F ) } addComment("[INFO]Expression data loaded and checked",T,opt$log,display=FALSE) addComment(c("[INFO]Dim of expression matrix:",dim(expDataMatrix)),T,opt$log,display=FALSE) #chargement du fichier des facteurs factorInfoMatrix=read.csv(file=file.path(getwd(), opt$factorInfo),header=F,sep="\t",colClasses="character") #remove first row to convert it as colnames colnames(factorInfoMatrix)=factorInfoMatrix[1,] factorInfoMatrix=factorInfoMatrix[-1,] #use first colum to convert it as rownames but not removing it to avoid conversion as vector in unique factor case rownames(factorInfoMatrix)=factorInfoMatrix[,1] if(length(setdiff(colnames(expDataMatrix),rownames(factorInfoMatrix)))!=0){ addComment("[ERROR]Missing samples in factor file",T,opt$log) q( "no", 1, F ) } #order sample as in expression matrix and remove spurious sample factorInfoMatrix=factorInfoMatrix[colnames(expDataMatrix),] #test if all values names are present in dico if(!all(unlist(factorInfoMatrix) %in% rownames(renamingDico))){ addComment("[ERROR]Missing factor names in renaming dictionary",T,opt$log) q( "no", 1, F ) } addComment("[INFO]Factors OK",T,opt$log,display=FALSE) addComment(c("[INFO]Dim of factorInfo matrix:",dim(factorInfoMatrix)),T,opt$log,display=FALSE) ##manage blocking factor blockingFactor=NULL blockinFactorsList=NULL if(!is.null(opt$blockingInfo)){ #chargement du fichier des blocking factors blockingInfoMatrix=read.csv(file=file.path(getwd(), opt$blockingInfo),header=F,sep="\t",colClasses="character") #remove first row to convert it as colnames colnames(blockingInfoMatrix)=blockingInfoMatrix[1,] blockingInfoMatrix=blockingInfoMatrix[-1,] #use first colum to convert it as rownames but not removing it to avoid conversion as vector in unique factor case rownames(blockingInfoMatrix)=blockingInfoMatrix[,1] if(length(setdiff(colnames(expDataMatrix),rownames(blockingInfoMatrix)))!=0){ addComment("[ERROR]Missing samples in blocking factor file",T,opt$log) q( "no", 1, F ) } #order sample as in expression matrix blockingInfoMatrix=blockingInfoMatrix[colnames(expDataMatrix),] #test if all blocking names are present in dico if(!all(unlist(blockingInfoMatrix) %in% rownames(renamingDico))){ addComment("[ERROR]Missing blocking names in renaming dictionary",T,opt$log) q( "no", 1, F ) } #remove blocking factors allready present as real factors blockingNotInMainFactors=setdiff(colnames(blockingInfoMatrix)[-1],colnames(factorInfoMatrix)[-1]) if(length(blockingNotInMainFactors)<(ncol(blockingInfoMatrix)-1))addComment("[WARNING]Blocking factors cannot be principal factors",T,opt$log,display=FALSE) if(length(blockingNotInMainFactors)>0){ blockingInfoMatrix=blockingInfoMatrix[,c(colnames(blockingInfoMatrix)[1],blockingNotInMainFactors)] groupBlocking=rep("c",ncol(expDataMatrix)) #for each blocking factor for(blockingFact in blockingNotInMainFactors){ if(opt$blockingPolicy=="correlated"){ indNewFact=as.numeric(factor(blockingInfoMatrix[,blockingFact])) groupBlocking=paste(groupBlocking,indNewFact,sep="_") }else{ if(is.null(blockinFactorsList))blockinFactorsList=list() blockinFactorsList[[blockingFact]]=factor(unlist(lapply(blockingInfoMatrix[,blockingFact],function(x)paste(c(blockingFact,"_",x),collapse="")))) } } if(opt$blockingPolicy=="correlated"){ blockingFactor=factor(groupBlocking) if(length(levels(blockingFactor))==1){ addComment("[ERROR]Selected blocking factors seems to be constant",T,opt$log) q( "no", 1, F ) } } addComment("[INFO]Blocking info OK",T,opt$log,display=FALSE) }else{ addComment("[WARNING]No blocking factors will be considered",T,opt$log,display=FALSE) } } ##rename different input parameters using renamingDictionary opt$factorsContrast=renamingDico[unlist(lapply(unlist(strsplit(opt$factorsContrast,",")),function(x)which(renamingDico[,1]==x))),2] userDefinedContrasts=FALSE if(!is.null(opt$firstGroupContrast) && !is.null(opt$secondGroupContrast)){ userDefinedContrasts=TRUE for(iContrast in 1:length(opt$firstGroupContrast)){ opt$firstGroupContrast[iContrast]=paste(unlist(lapply(unlist(strsplit(opt$firstGroupContrast[iContrast],",")),function(x)paste(renamingDico[unlist(lapply(unlist(strsplit(x,"\\*")),function(x)which(renamingDico[,1]==x))),2],collapse="*"))),collapse=",") opt$secondGroupContrast[iContrast]=paste(unlist(lapply(unlist(strsplit(opt$secondGroupContrast[iContrast],",")),function(x)paste(renamingDico[unlist(lapply(unlist(strsplit(x,"\\*")),function(x)which(renamingDico[,1]==x))),2],collapse="*"))),collapse=",") } } if(!is.null(opt$controlGroups)){ renamedGroups=c() for(iGroup in unlist(strsplit(opt$controlGroups,","))){ renamedControlGroup=paste(renamingDico[unlist(lapply(unlist(strsplit(iGroup,":")),function(x)which(renamingDico[,1]==x))),2],collapse=":") if(length(renamedControlGroup)==0 || any(which(unlist(gregexpr(text = renamedControlGroup,pattern = ":"))==-1))){ addComment("[ERROR]Control groups for interaction seem to mismatch, please check them.",T,opt$log) q( "no", 1, F ) } renamedGroups=c(renamedGroups,renamedControlGroup) } opt$controlGroups=renamedGroups } addComment("[INFO]Contrast variables are renamed to avoid confusion",T,opt$log,display=FALSE) ##renaming done #to convert factor as numeric value --> useless now ? #expDataMatrix=apply(expDataMatrix,c(1,2),function(x)as.numeric(paste(x))) #get factors info for LIMMA factorsList=list() for(iFactor in opt$factorsContrast){ if(!(iFactor %in% colnames(factorInfoMatrix))){ addComment("[ERROR]Required factors are missing in input file",T,opt$log) q( "no", 1, F ) } factorsList[[iFactor]]=factor(unlist(lapply(factorInfoMatrix[,iFactor],function(x)paste(c(iFactor,"_",x),collapse="")))) if(length(levels(factorsList[[iFactor]]))==1){ addComment("[ERROR]One selected factor seems to be constant",T,opt$log) q( "no", 1, F ) } } #check if there is at least 2 factors to allow interaction computation if(!is.null(opt$controlGroups) && length(factorsList)<2){ addComment("[ERROR]You cannot ask for interaction with less than 2 factors",T,opt$log) q( "no", 1, F ) } #merge all factors as a single one factorsMerged=as.character(factorsList[[opt$factorsContrast[1]]]) for(iFactor in opt$factorsContrast[-1]){ factorsMerged=paste(factorsMerged,as.character(factorsList[[iFactor]]),sep=".") } factorsMerged=factor(factorsMerged) #checked that coefficient number (ie. factorsMerged levels) is strictly smaller than sample size if(length(levels(factorsMerged))>=length(factorsMerged)){ addComment(c("[ERROR]No enough samples (",length(factorsMerged),") to estimate ",length(levels(factorsMerged))," coefficients"),T,opt$log) q( "no", 1, F ) } #get the sample size of each factor values sampleSizeFactor=table(factorsMerged) if(!is.null(blockinFactorsList)){ factorString=c("blockinFactorsList[['", names(blockinFactorsList)[1],"']]") for(blockingFact in names(blockinFactorsList)[-1]){ factorString=c(factorString," + blockinFactorsList[['",blockingFact,"']]") } design = model.matrix(as.formula(paste(c("~ factorsMerged +",factorString," + 0"),collapse=""))) #rename design columns coeffMeaning = levels(factorsMerged) for(blockingFact in blockinFactorsList){ coeffMeaning=c(coeffMeaning,levels(blockingFact)[-1]) } colnames(design) = coeffMeaning }else{ design = model.matrix(as.formula( ~ factorsMerged + 0)) #rename degin columns coeffMeaning = levels(factorsMerged) colnames(design) = coeffMeaning } addComment(c("[INFO]Available coefficients: ",coeffMeaning),T,opt$log,display=F) estimableCoeff=which(colSums(design)!=0) addComment("[INFO]Design done",T,opt$log,display=F) #use blocking factor if exists if(!is.null(blockingFactor)){ corfit <- duplicateCorrelation(expDataMatrix, design, block=blockingFactor) addComment(c("[INFO]Correlation within groups: ",corfit$consensus.correlation),T,opt$log,display=F) #run linear model fit data.fit = lmFit(expDataMatrix,design,block = blockingFactor, correlation=corfit$consensus.correlation) }else{ #run linear model fit data.fit = lmFit(expDataMatrix,design) } estimatedCoeff=which(!is.na(data.fit$coefficients[1,])) addComment("[INFO]Lmfit done",T,opt$log,display=F) #catch situation where some coefficients cannot be estimated, probably due to dependances between design columns #if(length(setdiff(estimableCoeff,estimatedCoeff))>0){ # addComment("[ERROR]Error in design matrix, check your group definitions",T,opt$log) # q( "no", 1, F ) #} #to strong condition, should return ERROR only when coefficients relative to principal factors cannot be estimated, otherwise, return a simple WARNING #define requested contrasts requiredContrasts=c() humanReadingContrasts=c() persoContrastName=c() if(userDefinedContrasts){ for(iContrast in 1:length(opt$firstGroupContrast)){ posGroup=unlist(lapply(unlist(strsplit(opt$firstGroupContrast[iContrast],",")),function(x)paste(paste(opt$factorsContrast,unlist(strsplit(x,"\\*")),sep="_"),collapse="."))) negGroup=unlist(lapply(unlist(strsplit(opt$secondGroupContrast[iContrast],",")),function(x)paste(paste(opt$factorsContrast,unlist(strsplit(x,"\\*")),sep="_"),collapse="."))) #clear posGroup and negGroup from empty groups emptyPosGroups=which(!(posGroup%in%coeffMeaning)) if(length(emptyPosGroups)>0){ addComment(c("[WARNING]The group(s)",posGroup[emptyPosGroups],"is/are removed from contrast as it/they is/are empty"),T,opt$log,display=FALSE) posGroup=posGroup[-emptyPosGroups] currentHumanContrast=paste(unlist(strsplit(opt$firstGroupContrast[iContrast],","))[-emptyPosGroups],collapse="+") }else{ currentHumanContrast=paste(unlist(strsplit(opt$firstGroupContrast[iContrast],",")),collapse="+") } emptyNegGroups=which(!(negGroup%in%coeffMeaning)) if(length(emptyNegGroups)>0){ addComment(c("[WARNING]The group(s)",negGroup[emptyNegGroups],"is/are removed from contrast as it/they is/are empty"),T,opt$log,display=FALSE) negGroup=negGroup[-emptyNegGroups] currentHumanContrast=paste(c(currentHumanContrast,unlist(strsplit(opt$secondGroupContrast[iContrast],","))[-emptyNegGroups]),collapse="-") }else{ currentHumanContrast=paste(c(currentHumanContrast,unlist(strsplit(opt$secondGroupContrast[iContrast],","))),collapse="-") } if(length(posGroup)==0 || length(negGroup)==0 ){ addComment(c("[WARNING]Contrast",currentHumanContrast,"cannot be estimated due to empty group"),T,opt$log,display=FALSE) }else{ if(all(posGroup%in%negGroup) && all(negGroup%in%posGroup)){ addComment(c("[WARNING]Contrast",currentHumanContrast,"cannot be estimated due to null contrast"),T,opt$log,display=FALSE) }else{ #get coefficients required for first group added as positive positiveCoeffWeights=sampleSizeFactor[posGroup]/sum(sampleSizeFactor[posGroup]) #positiveCoeffWeights=rep(1,length(posGroup)) #names(positiveCoeffWeights)=names(sampleSizeFactor[posGroup]) #get coefficients required for second group added as negative negativeCoeffWeights=sampleSizeFactor[negGroup]/sum(sampleSizeFactor[negGroup]) #negativeCoeffWeights=rep(1,length(negGroup)) #names(negativeCoeffWeights)=names(sampleSizeFactor[negGroup]) #build the resulting contrast currentContrast=paste(paste(positiveCoeffWeights[posGroup],posGroup,sep="*"),collapse="+") currentContrast=paste(c(currentContrast,paste(paste(negativeCoeffWeights[negGroup],negGroup,sep="*"),collapse="-")),collapse="-") requiredContrasts=c(requiredContrasts,currentContrast) #build the human reading contrast humanReadingContrasts=c(humanReadingContrasts,currentHumanContrast) if(!is.null(opt$contrastNames) && nchar(opt$contrastNames[iContrast])>0){ persoContrastName=c(persoContrastName,opt$contrastNames[iContrast]) }else{ persoContrastName=c(persoContrastName,"") } addComment(c("[INFO]Contrast added : ",currentHumanContrast),T,opt$log,display=F) addComment(c("with complete formula ",currentContrast),T,opt$log,display=F) } } } } #define the true formula with interactions to get interaction coefficients factorString=c("factorsList[['", names(factorsList)[1],"']]") for(iFactor in names(factorsList)[-1]){ factorString=c(factorString," * factorsList[['",iFactor,"']]") } if(!is.null(blockinFactorsList)){ for(blockingFact in names(blockinFactorsList)){ factorString=c(factorString," + blockinFactorsList[['",blockingFact,"']]") } } #should not be null at the end allFtestMeanSquare=NULL #to get the F-test values estimatedInteractions=rownames(anova(lm(as.formula(paste(c("expDataMatrix[1,] ~ ",factorString),collapse=""))))) estimatedInteractions=c(unlist(lapply(estimatedInteractions[-length(estimatedInteractions)],function(x){temp=unlist(strsplit(x,"[ \" | : ]"));paste(temp[seq(2,length(temp),3)],collapse=":")})),estimatedInteractions[length(estimatedInteractions)]) #rename estimated interaction terms using renamingDico estimatedInteractions=c(unlist(lapply(estimatedInteractions[-length(estimatedInteractions)],function(x)paste(renamingDico[unlist(strsplit(x,":")),1],collapse=":"))),estimatedInteractions[length(estimatedInteractions)]) t <- unlist(apply(expDataMatrix,1,function(x){temp=anova(lm(as.formula(paste(c("x ~ ",factorString),collapse=""))))$`Mean Sq`;temp/temp[length(temp)]})) allFtestMeanSquare <- t(matrix(t,nrow=length(estimatedInteractions))) #remove from allFtest rows containing NA if(length(which(is.na(allFtestMeanSquare[,1])))>0)allFtestMeanSquare=allFtestMeanSquare[-(which(is.na(allFtestMeanSquare[,1]))),] colnames(allFtestMeanSquare)=estimatedInteractions #add contrasts corresponding to interaction terms if(!is.null(opt$controlGroups)){ #first load user defined control group for each factor controlGroup=rep(NA,length(factorsList)) names(controlGroup)=names(factorsList) for(iGroup in opt$controlGroups){ splitGroup=unlist(strsplit(iGroup,":")) splitGroup[2]=paste(splitGroup[1],splitGroup[2],sep = "_") #check if defined control group is really a level of the corresponding factor if(!splitGroup[1]%in%names(controlGroup) || !splitGroup[2]%in%factorsList[[splitGroup[1]]]){ addComment(c("[ERROR]The factor name",splitGroup[1],"does not exist or group name",splitGroup[2]),T,opt$log) q( "no", 1, F ) } if(!is.na(controlGroup[splitGroup[1]])){ addComment("[ERROR]Several control groups are defined for the same factor, please select only one control group for each factor if you want to compute interaction contrasts",T,opt$log) q( "no", 1, F ) } controlGroup[splitGroup[1]]=splitGroup[2] } #check if all factor have a defined control group if(any(is.na(controlGroup))){ addComment("[ERROR]Missing control group for some factors, please check them if you want to compute interaction contrasts",T,opt$log) q( "no", 1, F ) } nbFactors=length(factorsList) interactionContrasts=c() contrastClass=c() #initialize list for the first level newPreviousLoopContrast=list() for(iFactorA in 1:(nbFactors-1)){ nameFactorA=names(factorsList)[iFactorA] compA=c() for(levelA in setdiff(levels(factorsList[[iFactorA]]),controlGroup[nameFactorA])){ compA=c(compA,paste(levelA,controlGroup[nameFactorA],sep="-")) } newPreviousLoopContrast[[as.character(iFactorA)]]=compA } #make a loop for growing interaction set for(globalIfactor in 1:(nbFactors-1)){ previousLoopContrast=newPreviousLoopContrast newPreviousLoopContrast=list() #factor A reuse contrasts made at previsous loop for(iFactorA in names(previousLoopContrast)){ compA=previousLoopContrast[[iFactorA]] if(max(as.integer(unlist(strsplit(iFactorA,"\\."))))<nbFactors){ #factor B is the new factor to include in intreraction set for(iFactorB in (max(as.integer(unlist(strsplit(iFactorA,"\\."))))+1):nbFactors){ nameFactorB=names(factorsList)[iFactorB] #keep contrasts identified trough interacting factors set newPreviousLoopContrast[[paste(iFactorA,iFactorB,sep=".")]]=c() for(iCompA in compA){ for(levelB in setdiff(levels(factorsList[[iFactorB]]),controlGroup[nameFactorB])){ #decompose the contrast compA to apply the new level of factor B on each term temp=unlist(strsplit(iCompA,"[ + ]")) splitCompA=temp[1] for(iTemp in temp[-1])splitCompA=c(splitCompA,"+",iTemp) splitCompA=unlist(lapply(splitCompA,function(x){temp=unlist(strsplit(x,"-"));splitCompB=temp[1];for(iTemp in temp[-1])splitCompB=c(splitCompB,"-",iTemp);splitCompB})) #apply on each contrast term the new level of factor B firstTerm=paste(unlist(lapply(splitCompA,function(x)if(x!="+" && x!="-"){paste(x,levelB,sep=".")}else{x})),collapse="") secondTerm=negativeExpression(paste(unlist(lapply(splitCompA,function(x)if(x!="+" && x!="-"){paste(x,controlGroup[nameFactorB],sep=".")}else{x})),collapse="")) currentContrast=paste(c(firstTerm,secondTerm),collapse="") newPreviousLoopContrast[[paste(iFactorA,iFactorB,sep=".")]]=c(newPreviousLoopContrast[[paste(iFactorA,iFactorB,sep=".")]],currentContrast) } } } } } for(iContrast in names(newPreviousLoopContrast)){ contrastClass=c(contrastClass,rep(iContrast,length(newPreviousLoopContrast[[iContrast]]))) } interactionContrasts=c(interactionContrasts,unlist(newPreviousLoopContrast)) } #make human title for interactions names(interactionContrasts)=contrastClass humanReadingInteraction=unlist(lapply(interactionContrasts,function(x)gsub("\\.",":",unlist(strsplit(x,"[+-]"))[1]))) contrastToIgnore=c() #complete with control groups and order to match to coeffs for(iContrast in 1:length(interactionContrasts)){ missingFactor=setdiff(1:nbFactors,as.integer(unlist(strsplit(names(interactionContrasts[iContrast]),"\\.")))) #decompose the contrast temp=unlist(strsplit(interactionContrasts[iContrast],"[ + ]")) splitContrast=temp[1] for(iTemp in temp[-1])splitContrast=c(splitContrast,"+",iTemp) splitContrast=unlist(lapply(splitContrast,function(x){temp=unlist(strsplit(x,"-"));splitCompB=temp[1];for(iTemp in temp[-1])splitCompB=c(splitCompB,"-",iTemp);splitCompB})) for(iFactor in missingFactor){ for(iTerm in 1:length(splitContrast)){ if(splitContrast[iTerm]!="+" && splitContrast[iTerm]!="-"){ splitTerm=unlist(strsplit(splitContrast[iTerm],"\\.")) if(iFactor==1)splitContrast[iTerm]=paste(c(controlGroup[names(factorsList)[iFactor]],splitTerm),collapse=".") if(iFactor==nbFactors)splitContrast[iTerm]=paste(c(splitTerm,controlGroup[names(factorsList)[iFactor]]),collapse=".") if(iFactor>1 && iFactor<nbFactors)splitContrast[iTerm]=paste(c(splitTerm[1:(iFactor-1)],controlGroup[names(factorsList)[iFactor]],splitTerm[iFactor:length(splitTerm)]),collapse=".") } } } interactionContrasts[iContrast]=paste(splitContrast,collapse="") if(all(splitContrast[seq(1,length(splitContrast),2)]%in%coeffMeaning)){ addComment(c("[INFO]Interaction contrast added : ",humanReadingInteraction[iContrast]),T,opt$log,display=F) addComment(c("with complete formula ",interactionContrasts[iContrast]),T,opt$log,display=F) }else{ contrastToIgnore=c(contrastToIgnore,iContrast) addComment(c("[WARNING]Interaction contrast",humanReadingInteraction[iContrast],"is removed due to empty group"),T,opt$log,display=F) } } #add interaction contrasts to global contrast list if(length(contrastToIgnore)>0){ requiredContrasts=c(requiredContrasts,interactionContrasts[-contrastToIgnore]) humanReadingContrasts=c(humanReadingContrasts,humanReadingInteraction[-contrastToIgnore]) persoContrastName=c(persoContrastName,rep("",length(humanReadingInteraction[-contrastToIgnore]))) }else{ requiredContrasts=c(requiredContrasts,interactionContrasts) humanReadingContrasts=c(humanReadingContrasts,humanReadingInteraction) persoContrastName=c(persoContrastName,rep("",length(humanReadingInteraction))) } }#end of intreaction contrasts #remove from requiredContrasts contrasts that cannot be estimated toRemove=unique(unlist(lapply(setdiff(coeffMeaning,names(estimatedCoeff)),function(x)grep(x,requiredContrasts)))) if(length(toRemove)>0){ addComment(c("[WARNING]",length(toRemove)," contrasts are removed, due to missing coefficients"),T,opt$log,display=FALSE) requiredContrasts=requiredContrasts[-toRemove] humanReadingContrasts=humanReadingContrasts[-toRemove] persoContrastName=persoContrastName[-toRemove] } if(length(requiredContrasts)==0){ addComment("[ERROR]No contrast to compute, please check your contrast definition.",T,opt$log) q( "no", 1, F ) } #compute for each contrast mean of coefficients in posGroup and negGroup for FC computation of log(FC) with LSmean as in Partek meanPosGroup=list() meanNegGroup=list() for(iContrast in 1:length(requiredContrasts)){ #define posGroup and negGroup #first split contrast temp=unlist(strsplit(requiredContrasts[iContrast],"[ + ]")) splitContrast=temp[1] for(iTemp in temp[-1])splitContrast=c(splitContrast,"+",iTemp) splitContrast=unlist(lapply(splitContrast,function(x){temp=unlist(strsplit(x,"-"));splitCompB=temp[1];for(iTemp in temp[-1])splitCompB=c(splitCompB,"-",iTemp);splitCompB})) #and then put each term in good group posGroup=c() negGroup=c() nextIsPos=TRUE for(iSplit in splitContrast){ if(iSplit=="+")nextIsPos=TRUE if(iSplit=="-")nextIsPos=FALSE if(iSplit!="-" && iSplit!="+"){ #remove weights of contrast terms iSplitBis=unlist(strsplit(iSplit,"[*]")) iSplitBis=iSplitBis[length(iSplitBis)] if(nextIsPos)posGroup=c(posGroup,iSplitBis) else negGroup=c(negGroup,iSplitBis) } } #compute means for each group meanPosGroup[[iContrast]]=apply(as.matrix(data.fit$coefficients[,posGroup],ncol=length(posGroup)),1,mean) meanNegGroup[[iContrast]]=apply(as.matrix(data.fit$coefficients[,negGroup],ncol=length(negGroup)),1,mean) } contrast.matrix = makeContrasts(contrasts=requiredContrasts,levels=design) data.fit.con = contrasts.fit(data.fit,contrast.matrix) addComment("[INFO]Contrast definition done",T,opt$log,T,display=FALSE) #compute LIMMA statistics data.fit.eb = eBayes(data.fit.con) addComment("[INFO]Estimation done",T,opt$log,T,display=FALSE) #adjust p.value through FDR data.fit.eb$adj_p.value=data.fit.eb$p.value for(iComparison in 1:ncol(data.fit.eb$adj_p.value)){ data.fit.eb$adj_p.value[,iComparison]=p.adjust(data.fit.eb$p.value[,iComparison],"fdr") } #add a new field based on LS-means for each contrast instead of global mean like they were calculated in coefficients field data.fit.eb$coefficientsLS=data.fit.eb$coefficients if(ncol(data.fit.eb$coefficients)!=length(meanPosGroup)){ addComment("[ERROR]Estimated contrasts number unexpected",T,opt$log) q( "no", 1, F ) } for(iContrast in 1:length(meanPosGroup)){ data.fit.eb$coefficientsLS[,iContrast]=meanPosGroup[[iContrast]][rownames(data.fit.eb$coefficientsLS)]-meanNegGroup[[iContrast]][rownames(data.fit.eb$coefficientsLS)] } #if requested replace coefficient computed as global mean by LS-means values if(useLSmean)data.fit.eb$coefficients=data.fit.eb$coefficientsLS addComment("[INFO]Core treatment done",T,opt$log,T,display=FALSE) ##convert humanReadingContrasts with namingDictionary to create humanReadingContrastsRenamed and keep original humanReadingContrasts names for file names humanReadingContrastsRenamed=rep("",length(humanReadingContrasts)) for(iContrast in 1:length(humanReadingContrasts)){ if(persoContrastName[iContrast]==""){ #if(verbose)addComment(humanReadingContrasts[iContrast]) specialCharacters=str_extract_all(humanReadingContrasts[iContrast],"[+|*|_|:|-]")[[1]] #if(verbose)addComment(specialCharacters) nameConverted=unlist(lapply(strsplit(humanReadingContrasts[iContrast],"[+|*|_|:|-]")[[1]],function(x)renamingDico[x,1])) #if(verbose)addComment(nameConverted) humanReadingContrastsRenamed[iContrast]=paste(nameConverted,specialCharacters,collapse="",sep="") #if(verbose)addComment(humanReadingContrastsRenamed[iContrast]) humanReadingContrastsRenamed[iContrast]=substr(humanReadingContrastsRenamed[iContrast],1,nchar(humanReadingContrastsRenamed[iContrast])-1) }else{ humanReadingContrastsRenamed[iContrast]=persoContrastName[iContrast] } } #write correspondances between plot file names (humanReadingContrasts) and displayed names in figure legends (humanReadingContrastsRenamed), usefull to define html items in xml file correspondanceTable=matrix("",ncol=2,nrow=ncol(data.fit.eb$p.value)) correspondanceTable[,1]=unlist(lapply(humanReadingContrasts,function(x)gsub(":","_INT_",gsub("\\+","_PLUS_",gsub("\\*","_AND_",x))))) correspondanceTable[,2]=humanReadingContrastsRenamed rownames(correspondanceTable)=correspondanceTable[,2] write.table(correspondanceTable,file=file.path(getwd(), "correspondanceFileNames.csv"),quote=FALSE,sep="\t",col.names = F,row.names = F) #plot nominal p-val histograms for selected comparisons histogramPerPage=6 if (!is.null(opt$histo)) { iToPlot=1 plotVector=list() nbComparisons=ncol(data.fit.eb$p.value) for (iComparison in 1:nbComparisons){ dataToPlot=data.frame(pval=data.fit.eb$p.value[,iComparison],id=rownames(data.fit.eb$p.value)) p <- ggplot(data=dataToPlot, aes(x=pval)) + geom_histogram(colour="red", fill="salmon") + theme_bw() + ggtitle(humanReadingContrastsRenamed[iComparison]) + ylab(label="Frequencies") + xlab(label="Nominal p-val") + theme(panel.border=element_blank(),plot.title = element_text(hjust = 0.5)) plotVector[[length(plotVector)+1]]=p pp <- ggplotly(p) htmlwidgets::saveWidget(as_widget(pp), paste(c(file.path(getwd(), "plotLyDir"),"/",opt$histo,"_",correspondanceTable[humanReadingContrastsRenamed[iComparison],1],".html"),collapse=""),selfcontained = F) if(iComparison==nbComparisons || length(plotVector)==histogramPerPage){ #plot and close the actual plot if(opt$format=="pdf"){ pdf(paste(c("./plotDir/",opt$histo,iToPlot,".pdf"),collapse=""))}else{ png(paste(c("./plotDir/",opt$histo,iToPlot,".png"),collapse="")) } multiplot(plotlist=plotVector,cols=2) dev.off() if(iComparison<nbComparisons){ #prepare for a new plotting file if necessary plotVector=list() iToPlot=iToPlot+1 } } } addComment("[INFO]Histograms drawn",T,opt$log,T,display=FALSE) } #plot F-test sum square barplot if(!is.null(allFtestMeanSquare)){ dataToPlot=data.frame(Fratio=apply(allFtestMeanSquare,2,mean),Factors=factor(colnames(allFtestMeanSquare),levels = colnames(allFtestMeanSquare))) p <- ggplot(data=dataToPlot, aes(x=Factors, y=Fratio, fill=Factors)) + geom_bar(stat="identity") + scale_fill_manual(values = colorRampPalette(brewer.pal(9,"Set1"))(ncol(allFtestMeanSquare))[sample(ncol(allFtestMeanSquare))]) + ylab(label="mean F-ratio") + theme_bw() + theme(panel.border=element_blank(),plot.title = element_text(hjust = 0.5)) + ggtitle("Source of variation") if(opt$format=="pdf"){ pdf(paste(c("./plotDir/",opt$fratioFile,".pdf"),collapse=""))}else{ png(paste(c("./plotDir/",opt$fratioFile,".png"),collapse="")) } plot(p) dev.off() pp <- ggplotly(p) htmlwidgets::saveWidget(as_widget(pp), paste(c(file.path(getwd(), "plotLyDir"),"/",opt$fratioFile,".html"),collapse=""),selfcontained = F) addComment("[INFO]SumSquareTest drawn",T,opt$log,T,display=FALSE) } #plot VOLCANO plot #volcanoplot(data.fit.eb,coef=1,highlight=10) volcanoPerPage=1 logFCthreshold=log2(opt$thresholdFC) if (!is.null(opt$volcano)) { iToPlot=1 plotVector=list() nbComparisons=ncol(data.fit.eb$adj_p.value) for (iComparison in 1:nbComparisons){ #define the log10(p-val) threshold corresponding to FDR threshold fixed by user probeWithLowFDR=-log10(data.fit.eb$p.value[which(data.fit.eb$adj_p.value[,iComparison]<=opt$fdrThreshold),iComparison]) pvalThresholdFDR=NULL if(length(probeWithLowFDR)>0)pvalThresholdFDR=min(probeWithLowFDR) #get significant points over FC and FDR thresholds significativePoints=intersect(which(abs(data.fit.eb$coefficients[,iComparison])>=logFCthreshold),which(data.fit.eb$adj_p.value[,iComparison]<=opt$fdrThreshold)) #to reduce size of html plot, we keep 20000 points maximum sampled amongst genes with pval>=33%(pval) and abs(log2(FC))<=66%(abs(log2(FC))) htmlPointsToRemove=intersect(which(abs(data.fit.eb$coefficients[,iComparison])<=quantile(abs(data.fit.eb$coefficients[,iComparison]),c(0.66))),which(data.fit.eb$p.value[,iComparison]>=quantile(abs(data.fit.eb$p.value[,iComparison]),c(0.33)))) if(length(htmlPointsToRemove)>20000){ htmlPointsToRemove=setdiff(htmlPointsToRemove,sample(htmlPointsToRemove,20000)) }else{ htmlPointsToRemove=c() } xMinLimPlot=min(data.fit.eb$coefficients[,iComparison])-0.2 xMaxLimPlot=max(data.fit.eb$coefficients[,iComparison])+0.2 yMaxLimPlot= max(-log10(data.fit.eb$p.value[,iComparison]))+0.2 if(length(significativePoints)>0){ dataSignifToPlot=data.frame(pval=-log10(data.fit.eb$p.value[significativePoints,iComparison]),FC=data.fit.eb$coefficients[significativePoints,iComparison],description=paste(names(data.fit.eb$coefficients[significativePoints,iComparison]),"\n","FC: " , round(2^data.fit.eb$coefficients[significativePoints,iComparison],2) , " | FDR p-val: ",prettyNum(data.fit.eb$adj_p.value[significativePoints,iComparison],digits=4), sep="")) #to test if remains any normal points to draw if(length(significativePoints)<nrow(data.fit.eb$p.value)){ dataToPlot=data.frame(pval=-log10(data.fit.eb$p.value[-significativePoints,iComparison]),FC=data.fit.eb$coefficients[-significativePoints,iComparison],description=paste("FC: " , round(2^data.fit.eb$coefficients[-significativePoints,iComparison],2) , " | FDR p-val: ",prettyNum(data.fit.eb$adj_p.value[-significativePoints,iComparison],digits=4), sep="")) }else{ dataToPlot=data.frame(pval=0,FC=0,description="null") } }else{ dataToPlot=data.frame(pval=-log10(data.fit.eb$p.value[,iComparison]),FC=data.fit.eb$coefficients[,iComparison],description=paste("FC: " , round(2^data.fit.eb$coefficients[,iComparison],2) , " | FDR p-val: ",prettyNum(data.fit.eb$adj_p.value[,iComparison],digits=4), sep="")) } ##traditional plot p <- ggplot(data=dataToPlot, aes(x=FC, y=pval)) + geom_point() + theme_bw() + ggtitle(humanReadingContrastsRenamed[iComparison]) + ylab(label="-log10(p-val)") + xlab(label="Log2 Fold Change") + theme(panel.border=element_blank(),plot.title = element_text(hjust = 0.5),legend.position="none") if(logFCthreshold!=0) p <- p + geom_vline(xintercept=-logFCthreshold, color="salmon",linetype="dotted", size=1) + geom_vline(xintercept=logFCthreshold, color="salmon",linetype="dotted", size=1) + geom_text(data.frame(text=c(paste(c("log2(1/FC=",opt$thresholdFC,")"),collapse=""),paste(c("log2(FC=",opt$thresholdFC,")"),collapse="")),x=c(-logFCthreshold,logFCthreshold),y=c(0,0)),mapping=aes(x=x, y=y, label=text), size=4, angle=90, vjust=-0.4, hjust=0, color="salmon") if(!is.null(pvalThresholdFDR)) p <- p + geom_hline(yintercept=pvalThresholdFDR, color="skyblue1",linetype="dotted", size=0.5) + geom_text(data.frame(text=c(paste(c("FDR pval limit(",opt$fdrThreshold,")"),collapse="")),x=c(xMinLimPlot),y=c(pvalThresholdFDR)),mapping=aes(x=x, y=y, label=text), size=4, vjust=0, hjust=0, color="skyblue3") if(length(significativePoints)>0)p <- p + geom_point(data=dataSignifToPlot,aes(colour=description)) ##interactive plot if(length(htmlPointsToRemove)>0){ pointToRemove=union(htmlPointsToRemove,significativePoints) #to test if remains any normal points to draw if(length(pointToRemove)<nrow(data.fit.eb$p.value)){ dataToPlot=data.frame(pval=-log10(data.fit.eb$p.value[-pointToRemove,iComparison]),FC=data.fit.eb$coefficients[-pointToRemove,iComparison],description=paste("FC: " , round(2^data.fit.eb$coefficients[-pointToRemove,iComparison],2) , " | FDR p-val: ", prettyNum(data.fit.eb$adj_p.value[-pointToRemove,iComparison],digits=4), sep="")) }else{ dataToPlot=data.frame(pval=0,FC=0,description="null") } } if((nrow(dataToPlot)+nrow(dataSignifToPlot))>40000)addComment(c("[WARNING]For",humanReadingContrastsRenamed[iComparison],"volcano, numerous points to plot(",nrow(dataToPlot)+nrow(dataSignifToPlot),"), resulting volcano could be heavy, using more stringent thresholds could be helpful."),T,opt$log) phtml <- plot_ly(data=dataToPlot, x=~FC, y=~pval,type="scatter", mode="markers",showlegend = FALSE, marker = list(color="gray",opacity=0.5), text=~description, hoverinfo="text") %>% layout(title = humanReadingContrastsRenamed[iComparison],xaxis=list(title="Log2 Fold Change",showgrid=TRUE, zeroline=FALSE),yaxis=list(title="-log10(p-val)", showgrid=TRUE, zeroline=FALSE)) if(length(significativePoints)>0) phtml=add_markers(phtml,data=dataSignifToPlot, x=~FC, y=~pval, mode="markers" , marker=list( color=log10(abs(dataSignifToPlot$FC)*dataSignifToPlot$pval),colorscale='Rainbow'), text=~description, hoverinfo="text", inherit = FALSE) %>% hide_colorbar() if(logFCthreshold!=0){ phtml=add_trace(phtml,x=c(-logFCthreshold,-logFCthreshold), y=c(0,yMaxLimPlot), type="scatter", mode = "lines", line=list(color="coral",dash="dash"), hoverinfo='none', showlegend = FALSE,inherit = FALSE) phtml=add_annotations(phtml,x=-logFCthreshold,y=0,xref = "x",yref = "y",text = paste(c("log2(1/FC=",opt$thresholdFC,")"),collapse=""),xanchor = 'right',showarrow = F,textangle=270,font=list(color="coral")) phtml=add_trace(phtml,x=c(logFCthreshold,logFCthreshold), y=c(0, yMaxLimPlot), type="scatter", mode = "lines", line=list(color="coral",dash="dash"), hoverinfo='none', showlegend = FALSE,inherit = FALSE) phtml=add_annotations(phtml,x=logFCthreshold,y=0,xref = "x",yref = "y",text = paste(c("log2(FC=",opt$thresholdFC,")"),collapse=""),xanchor = 'right',showarrow = F,textangle=270,font=list(color="coral")) } if(!is.null(pvalThresholdFDR)){ phtml=add_trace(phtml,x=c(xMinLimPlot,xMaxLimPlot), y=c(pvalThresholdFDR,pvalThresholdFDR), type="scatter", mode = "lines", line=list(color="cornflowerblue",dash="dash"), hoverinfo='none', showlegend = FALSE,inherit = FALSE) phtml=add_annotations(phtml,x=xMinLimPlot,y=pvalThresholdFDR+0.1,xref = "x",yref = "y",text = paste(c("FDR pval limit(",opt$fdrThreshold,")"),collapse=""),xanchor = 'left',showarrow = F,font=list(color="cornflowerblue")) } plotVector[[length(plotVector)+1]]=p #save plotly files pp <- ggplotly(phtml) htmlwidgets::saveWidget(as_widget(pp), paste(c(file.path(getwd(), "plotLyDir"),"/",opt$volcano,"_",correspondanceTable[humanReadingContrastsRenamed[iComparison],1],".html"),collapse=""),selfcontained = F) if(iComparison==nbComparisons || length(plotVector)==volcanoPerPage){ #plot and close the actual plot if(opt$format=="pdf"){ pdf(paste(c("./plotDir/",opt$volcano,"_",correspondanceTable[humanReadingContrastsRenamed[iComparison],1],".pdf"),collapse=""))}else{ png(paste(c("./plotDir/",opt$volcano,"_",correspondanceTable[humanReadingContrastsRenamed[iComparison],1],".png"),collapse="")) } multiplot(plotlist=plotVector,cols=1) dev.off() if(iComparison<nbComparisons){ #prepare for a new ploting file if necessary plotVector=list() iToPlot=iToPlot+1 } } } remove(dataToPlot,dataSignifToPlot) addComment("[INFO]Volcanos drawn",T,opt$log,T,display=FALSE) } rowItemInfo=NULL if(!is.null(opt$rowNameType) && !is.null(opt$organismID)){ ##get gene information from BioMart #if(!require("biomaRt")){ # source("https://bioconductor.org/biocLite.R") # biocLite("biomaRt") #} ensembl_hs_mart <- useMart(biomart="ensembl", dataset=opt$organismID) ensembl_df <- getBM(attributes=c(opt$rowNameType,"description"),mart=ensembl_hs_mart) rowItemInfo=ensembl_df[which(ensembl_df[,1]!=""),2] rowItemInfo=unlist(lapply(rowItemInfo,function(x)substr(unlist(strsplit(x," \\[Source"))[1],1,30))) names(rowItemInfo)=ensembl_df[which(ensembl_df[,1]!=""),1] } #write(unlist(dimnames(data.fit.eb$adj_p.value)),opt$log,append = T) #prepare additional output containing df informations dfMatrix=matrix(0,ncol=3,nrow = nrow(data.fit.eb$coefficients),dimnames = list(rownames(data.fit.eb$coefficients),c("df.residual","df.prior","df.total"))) dfMatrix[,"df.residual"]=data.fit.eb$df.residual dfMatrix[,"df.prior"]=data.fit.eb$df.prior dfMatrix[,"df.total"]=data.fit.eb$df.total #filter out genes with higher p-values for all comparisons genesToKeep=names(which(apply(data.fit.eb$adj_p.value,1,function(x)length(which(x<=opt$fdrThreshold))>0))) #filter out genes with lower FC for all comparisons genesToKeep=intersect(genesToKeep,names(which(apply(data.fit.eb$coefficients,1,function(x)length(which(abs(x)>=logFCthreshold))>0)))) if(length(genesToKeep)>0){ data.fit.eb$adj_p.value=matrix(data.fit.eb$adj_p.value[genesToKeep,],ncol=ncol(data.fit.eb$adj_p.value)) rownames(data.fit.eb$adj_p.value)=genesToKeep colnames(data.fit.eb$adj_p.value)=colnames(data.fit.eb$p.value) data.fit.eb$p.value=matrix(data.fit.eb$p.value[genesToKeep,],ncol=ncol(data.fit.eb$p.value)) rownames(data.fit.eb$p.value)=genesToKeep colnames(data.fit.eb$p.value)=colnames(data.fit.eb$adj_p.value) data.fit.eb$coefficients=matrix(data.fit.eb$coefficients[genesToKeep,],ncol=ncol(data.fit.eb$coefficients)) rownames(data.fit.eb$coefficients)=genesToKeep colnames(data.fit.eb$coefficients)=colnames(data.fit.eb$adj_p.value) data.fit.eb$t=matrix(data.fit.eb$t[genesToKeep,],ncol=ncol(data.fit.eb$t)) rownames(data.fit.eb$t)=genesToKeep colnames(data.fit.eb$t)=colnames(data.fit.eb$adj_p.value) dfMatrix=dfMatrix[genesToKeep,,drop=FALSE] }else{ addComment(c("[WARNING]No significative genes considering the given FDR threshold : ",opt$fdrThreshold),T,opt$log,display=FALSE) } addComment("[INFO]Significant genes filtering done",T,opt$log,T,display=FALSE) #plot VennDiagramm for genes below threshold between comparisons #t=apply(data.fit.eb$adj_p.value[,1:4],2,function(x)names(which(x<=opt$threshold))) #get.venn.partitions(t) #vennCounts(data.fit.eb$adj_p.value[,1:4]<=opt$threshold) #make a simple sort genes based only on the first comparison #newOrder=order(data.fit.eb$adj_p.value[,1]) #data.fit.eb$adj_p.value=data.fit.eb$adj_p.value[newOrder,] #alternative sorting strategy based on the mean gene rank over all comparisons if(length(genesToKeep)>1){ currentRank=rep(0,nrow(data.fit.eb$adj_p.value)) for(iComparison in 1:ncol(data.fit.eb$adj_p.value)){ currentRank=currentRank+rank(data.fit.eb$adj_p.value[,iComparison]) } currentRank=currentRank/ncol(data.fit.eb$adj_p.value) newOrder=order(currentRank) data.fit.eb$adj_p.value=matrix(data.fit.eb$adj_p.value[newOrder,],ncol=ncol(data.fit.eb$adj_p.value)) rownames(data.fit.eb$adj_p.value)=rownames(data.fit.eb$p.value)[newOrder] colnames(data.fit.eb$adj_p.value)=colnames(data.fit.eb$p.value) data.fit.eb$p.value=matrix(data.fit.eb$p.value[newOrder,],ncol=ncol(data.fit.eb$p.value)) rownames(data.fit.eb$p.value)=rownames(data.fit.eb$adj_p.value) colnames(data.fit.eb$p.value)=colnames(data.fit.eb$adj_p.value) data.fit.eb$coefficients=matrix(data.fit.eb$coefficients[newOrder,],ncol=ncol(data.fit.eb$coefficients)) rownames(data.fit.eb$coefficients)=rownames(data.fit.eb$adj_p.value) colnames(data.fit.eb$coefficients)=colnames(data.fit.eb$adj_p.value) data.fit.eb$t=matrix(data.fit.eb$t[newOrder,],ncol=ncol(data.fit.eb$t)) rownames(data.fit.eb$t)=rownames(data.fit.eb$adj_p.value) colnames(data.fit.eb$t)=colnames(data.fit.eb$adj_p.value) dfMatrix=dfMatrix[newOrder,,drop=FALSE] } #formating output matrices depending on genes to keep if(length(genesToKeep)==0){ outputData=matrix(0,ncol=ncol(data.fit.eb$adj_p.value)*5+2,nrow=3) outputData[1,]=c("X","X",rep(humanReadingContrastsRenamed,each=5)) outputData[2,]=c("X","X",rep(c("p-val","FDR.p-val","FC","log2(FC)","t-stat"),ncol(data.fit.eb$adj_p.value))) outputData[,1]=c("LIMMA","Gene","noGene") outputData[,2]=c("Comparison","Info","noInfo") outputDfData=matrix(0,ncol=3+1,nrow=2) outputDfData[1,]=c("X","df.residual","df.prior","df.total") outputDfData[,1]=c("Statistics","noGene") }else{ if(length(genesToKeep)==1){ outputData=matrix(0,ncol=ncol(data.fit.eb$adj_p.value)*5+2,nrow=3) outputData[1,]=c("X","X",rep(humanReadingContrastsRenamed,each=5)) outputData[2,]=c("X","X",rep(c("p-val","FDR.p-val","FC","log2(FC)","t-stat"),ncol(data.fit.eb$adj_p.value))) outputData[,1]=c("LIMMA","Gene",genesToKeep) outputData[,2]=c("Comparison","Info","na") if(!is.null(rowItemInfo))outputData[3,2]=rowItemInfo[genesToKeep] outputData[3,seq(3,ncol(outputData),5)]=prettyNum(data.fit.eb$p.value,digits=4) outputData[3,seq(4,ncol(outputData),5)]=prettyNum(data.fit.eb$adj_p.value,digits=4) outputData[3,seq(5,ncol(outputData),5)]=prettyNum(2^data.fit.eb$coefficients,digits=4) outputData[3,seq(6,ncol(outputData),5)]=prettyNum(data.fit.eb$coefficients,digits=4) outputData[3,seq(7,ncol(outputData),5)]=prettyNum(data.fit.eb$t,digits=4) outputDfData=matrix(0,ncol=3+1,nrow=1+nrow(dfMatrix)) outputDfData[1,]=c("Statistics","df.residual","df.prior","df.total") outputDfData[2,]=c(rownames(dfMatrix),prettyNum(dfMatrix[,c("df.residual","df.prior","df.total")],digits=4)) }else{ #format matrix to be correctly read by galaxy (move headers in first column and row) outputData=matrix(0,ncol=ncol(data.fit.eb$adj_p.value)*5+2,nrow=nrow(data.fit.eb$adj_p.value)+2) outputData[1,]=c("X","X",rep(humanReadingContrastsRenamed,each=5)) outputData[2,]=c("X","X",rep(c("p-val","FDR.p-val","FC","log2(FC)","t-stat"),ncol(data.fit.eb$adj_p.value))) outputData[,1]=c("LIMMA","Gene",rownames(data.fit.eb$adj_p.value)) outputData[,2]=c("Comparison","Info",rep("na",nrow(data.fit.eb$adj_p.value))) if(!is.null(rowItemInfo))outputData[3:nrow(outputData),2]=rowItemInfo[rownames(data.fit.eb$adj_p.value)] outputData[3:nrow(outputData),seq(3,ncol(outputData),5)]=prettyNum(data.fit.eb$p.value,digits=4) outputData[3:nrow(outputData),seq(4,ncol(outputData),5)]=prettyNum(data.fit.eb$adj_p.value,digits=4) outputData[3:nrow(outputData),seq(5,ncol(outputData),5)]=prettyNum(2^data.fit.eb$coefficients,digits=4) outputData[3:nrow(outputData),seq(6,ncol(outputData),5)]=prettyNum(data.fit.eb$coefficients,digits=4) outputData[3:nrow(outputData),seq(7,ncol(outputData),5)]=prettyNum(data.fit.eb$t,digits=4) outputDfData=matrix(0,ncol=3+1,nrow=1+nrow(dfMatrix)) outputDfData[1,]=c("Statistics","df.residual","df.prior","df.total") outputDfData[2:(1+nrow(dfMatrix)),]=cbind(rownames(dfMatrix),prettyNum(dfMatrix[,c("df.residual")],digits=4),prettyNum(dfMatrix[,c("df.prior")],digits=4),prettyNum(dfMatrix[,c("df.total")],digits=4)) } } addComment("[INFO]Formated output",T,opt$log,display=FALSE) #write output results write.table(outputData,file=opt$outputFile,quote=FALSE,sep="\t",col.names = F,row.names = F) #write df info file write.table(outputDfData,file=opt$outputDfFile,quote=FALSE,sep="\t",col.names = F,row.names = F) end.time <- Sys.time() addComment(c("[INFO]Total execution time for R script:",as.numeric(end.time - start.time,units="mins"),"mins"),T,opt$log,display=FALSE) addComment("[INFO]End of R script",T,opt$log,display=FALSE) printSessionInfo(opt$log) #sessionInfo()