diff src/heatMapClustering.R @ 0:3022feec50fe draft

"planemo upload for repository https://github.com/juliechevalier/GIANT/tree/master commit cb276a594444c8f32e9819fefde3a21f121d35df"
author vandelj
date Fri, 26 Jun 2020 09:36:46 -0400
parents
children
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/src/heatMapClustering.R	Fri Jun 26 09:36:46 2020 -0400
@@ -0,0 +1,896 @@
+# A command-line interface to plot heatmap based on expression or diff. exp. analysis 
+# 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, OutDec=".")
+
+#get options
+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(
+  "expressionFile", "x", 1, "character",
+  "diffAnalyseFile", "x", 1, "character",
+  "factorInfo","x", 1, "character",
+  "genericData","x", 0, "logical",
+  "comparisonName","x",1,"character",
+  "comparisonNameLow","x",1,"character",
+  "comparisonNameHigh","x",1,"character",
+  "filterInputOutput","x", 1, "character",
+  "FCthreshold","x", 1, "double",
+  "pvalThreshold","x", 1, "double",
+  "geneListFiltering","x",1,"character",
+  "clusterNumber","x",1,"integer",
+  "maxRows","x",1,"integer",
+  "sampleClusterNumber","x",1,"integer",
+  "dataTransformation","x",1,"character",
+  "distanceMeasure","x",1,"character",
+  "aggloMethod","x",1,"character",
+  "personalColors","x",1,"character",
+  "sideBarColorPalette","x",1,"character",
+  "format", "x", 1, "character",
+  "quiet", "x", 0, "logical",
+  "log", "x", 1, "character",
+  "outputFile" , "x", 1, "character"),
+  byrow=TRUE, ncol=4)
+opt <- getoptLong(spec)
+
+# enforce the following required arguments
+if (is.null(opt$log)) {
+  addComment("[ERROR]'log file' is required")
+  q( "no", 1, F )
+}
+addComment("[INFO]Start of R script",T,opt$log,display=FALSE)
+if (is.null(opt$format)) {
+  addComment("[ERROR]'output format' is required",T,opt$log)
+  q( "no", 1, F )
+}
+if (is.null(opt$outputFile)) {
+  addComment("[ERROR]'output file' is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if(is.null(opt$expressionFile) && !is.null(opt$genericData)){
+  addComment("[ERROR]generic data clustering is based on expression clustering",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$clusterNumber) || opt$clusterNumber<2) {
+  addComment("[ERROR]valid genes clusters number is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$sampleClusterNumber) || opt$sampleClusterNumber<1) {
+  addComment("[ERROR]valid samples clusters number is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$dataTransformation)) {
+  addComment("[ERROR]data transformation option is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$distanceMeasure)) {
+  addComment("[ERROR]distance measure option is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$aggloMethod)) {
+  addComment("[ERROR]agglomeration method option is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$maxRows) || opt$maxRows<2) {
+  addComment("[ERROR]valid plotted row number is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (!is.null(opt[["comparisonName"]]) && nchar(opt[["comparisonName"]])==0){
+  addComment("[ERROR]you have to specify comparison",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (!is.null(opt$comparisonNameLow) && nchar(opt$comparisonNameLow)==0){
+  addComment("[ERROR]you have to specify comparisonLow",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (!is.null(opt$comparisonNameHigh) && nchar(opt$comparisonNameHigh)==0){
+  addComment("[ERROR]you have to specify comparisonHigh",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (is.null(opt$genericData) && (!is.null(opt$comparisonNameLow) || !is.null(opt$comparisonNameHigh))){
+  addComment("[ERROR]comparisonLow and comparisonHigh can be specified only with generic data",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (!is.null(opt$genericData) && !is.null(opt[["comparisonName"]])){
+  addComment("[ERROR]basic comparison cannot be specified for generic data",T,opt$log)
+  q( "no", 1, F )
+}
+
+if ((!is.null(opt[["comparisonName"]]) || !is.null(opt$comparisonNameLow) || !is.null(opt$comparisonNameHigh)) && is.null(opt$diffAnalyseFile)) {
+  addComment("[ERROR]'diff. exp. analysis file' is required",T,opt$log)
+  q( "no", 1, F )
+}
+
+if (!is.null(opt$genericData) && !is.null(opt$diffAnalyseFile) && is.null(opt$comparisonNameLow) && is.null(opt$comparisonNameHigh)){
+  addComment("[ERROR]Missing comparison information for filtering",T,opt$log)
+  q( "no", 1, F )
+}
+
+if ((!is.null(opt$FCthreshold) || !is.null(opt$pvalThreshold)) && (is.null(opt[["comparisonName"]]) && is.null(opt$comparisonNameLow) && is.null(opt$comparisonNameHigh))) {
+  addComment("[ERROR]'comparisons' are missing for filtering",T,opt$log)
+  q( "no", 1, F )
+}
+
+if ((!is.null(opt$FCthreshold) || !is.null(opt$pvalThreshold)) && !is.null(opt$geneListFiltering)) {
+  addComment("[ERROR]Cannot have two filtering strategies",T,opt$log)
+  q( "no", 1, F )
+}
+
+verbose <- if (is.null(opt$quiet)) {
+  TRUE
+}else{
+  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$log,display=FALSE)
+
+#directory for plots and HTML
+dir.create(file.path(getwd(), "plotDir"))
+dir.create(file.path(getwd(), "plotLyDir"))
+
+#silent package loading
+suppressPackageStartupMessages({
+  library("plotly")
+  library("dendextend")
+  #library("ggdendro")
+  #library("plyr")
+  library("ggplot2")
+  library("heatmaply")
+  library("circlize")
+  #library("RColorBrewer")
+  #source("https://bioconductor.org/biocLite.R")
+  #biocLite("ComplexHeatmap")
+  library("ComplexHeatmap")
+  #library("processx")
+})
+
+expressionToCluster=!is.null(opt$expressionFile)
+
+#load input data files
+if(expressionToCluster){
+  #first expression data
+  expressionMatrix=read.csv(file=opt$expressionFile,header=F,sep="\t",colClasses="character")
+  #remove first row to convert it as colnames (to avoid X before colnames with header=T)
+  colNamesData=expressionMatrix[1,-1]
+  expressionMatrix=expressionMatrix[-1,]
+  #remove first colum to convert it as rownames
+  rowNamesData=expressionMatrix[,1]
+  expressionMatrix=expressionMatrix[,-1]
+  if(is.data.frame(expressionMatrix)){
+    expressionMatrix=data.matrix(expressionMatrix)
+  }else{
+    expressionMatrix=data.matrix(as.numeric(expressionMatrix))
+  }
+  dimnames(expressionMatrix)=list(rowNamesData,colNamesData)
+  
+  #check input files
+  if (!is.numeric(expressionMatrix)) {
+    addComment("[ERROR]Expression data is not fully numeric!",T,opt$log,display=FALSE)
+    q( "no", 1, F )
+  }
+  
+  addComment("[INFO]Expression data loaded and checked")
+  addComment(c("[INFO]Dim of expression matrix:",dim(expressionMatrix)),T,opt$log,display=FALSE)
+}
+
+nbComparisons=0
+nbColPerContrast=5
+comparisonMatrix=NULL
+comparisonMatrixInfoGene=NULL
+#if available comparisons
+if(!is.null(opt[["comparisonName"]])){
+    #load results from differential expression analysis
+    #consider first row contains column names
+    comparisonMatrix=read.csv(file=opt$diffAnalyseFile,header=F,sep="\t")
+    colnames(comparisonMatrix)=as.character(unlist(comparisonMatrix[1,]))
+    #remove the second line also as it's information line (p-val,FDR.p-val,FC,logFC)
+    comparisonMatrix=comparisonMatrix[-c(1,2),]
+    #remove first and second colums, convert the first one as rownames
+    rownames(comparisonMatrix)=as.character(unlist(comparisonMatrix[,1]))
+    #and save second column content that contain geneInfo
+    comparisonMatrixInfoGene=as.character(unlist(comparisonMatrix[,2]))
+    names(comparisonMatrixInfoGene)=as.character(unlist(comparisonMatrix[,1]))
+    comparisonMatrix=comparisonMatrix[,-c(1,2)]
+    
+    comparisonMatrix=matrix(as.numeric(as.matrix(comparisonMatrix)),ncol=ncol(comparisonMatrix),dimnames = dimnames(comparisonMatrix))
+    
+    if (ncol(comparisonMatrix)%%nbColPerContrast != 0) {
+      addComment("[ERROR]Diff. exp. data does not contain good number of columns per contrast, should contains in this order:p-val,FDR.p-val,FC,log2(FC) and t-stat",T,opt$log,display=FALSE)
+      q( "no", 1, F )
+    }
+    
+    if(max(comparisonMatrix[,c(seq(1,ncol(comparisonMatrix),nbColPerContrast),seq(2,ncol(comparisonMatrix),nbColPerContrast))])>1 || min(comparisonMatrix[,c(seq(1,ncol(comparisonMatrix),nbColPerContrast),seq(2,ncol(comparisonMatrix),nbColPerContrast))])<0){
+      addComment("[ERROR]Seem that diff. exp. data does not contain correct values for p-val and FDR.p-val columns, should be including in [0,1] interval",T,opt$log,display=FALSE)
+      q( "no", 1, F )
+    }
+    
+    if (!is.numeric(comparisonMatrix)) {
+      addComment("[ERROR]Diff. exp. data is not fully numeric!",T,opt$log,display=FALSE)
+      q( "no", 1, F )
+    }
+    
+    if(expressionToCluster && length(setdiff(rownames(comparisonMatrix),rownames(expressionMatrix)))!=0){
+      addComment("[WARNING]All genes from diff. exp. file are not included in expression file",T,opt$log,display=FALSE)
+    }
+    
+    if(expressionToCluster && length(setdiff(rownames(expressionMatrix),rownames(comparisonMatrix)))!=0){
+      addComment("[WARNING]All genes from expression file are not included in diff. exp. file",T,opt$log,display=FALSE)
+    }
+    
+    addComment("[INFO]Diff. exp. analysis loaded and checked",T,opt$log,display=FALSE)
+    addComment(c("[INFO]Dim of original comparison matrix:",dim(comparisonMatrix)),T,opt$log,display=FALSE)
+    
+    #restrict to user specified comparisons
+    restrictedComparisons=unlist(strsplit(opt[["comparisonName"]],","))
+    #should be improved to avoid selection of column names starting too similarly  
+    colToKeep=which(unlist(lapply(colnames(comparisonMatrix),function(x)any(startsWith(x,restrictedComparisons)))))
+    comparisonMatrix=matrix(comparisonMatrix[,colToKeep],ncol=length(colToKeep),dimnames = list(rownames(comparisonMatrix),colnames(comparisonMatrix)[colToKeep]))
+    
+    #get number of required comparisons
+    nbComparisons=ncol(comparisonMatrix)/nbColPerContrast
+    
+    addComment(c("[INFO]Dim of effective filtering matrix:",dim(comparisonMatrix)),T,opt$log,display=FALSE)
+}
+
+#should be only the case with generic data
+if(!is.null(opt$comparisonNameLow) || !is.null(opt$comparisonNameHigh)){
+    #load generic data used for filtering
+    nbColPerContrast=1
+    #consider first row contains column names
+    comparisonMatrix=read.csv(file=opt$diffAnalyseFile,header=F,sep="\t")
+    colnames(comparisonMatrix)=as.character(unlist(comparisonMatrix[1,]))
+    #remove first colum, convert the first one as rownames
+    rownames(comparisonMatrix)=as.character(unlist(comparisonMatrix[,1]))
+    comparisonMatrix=comparisonMatrix[-1,-1]
+    
+    comparisonMatrix=matrix(as.numeric(as.matrix(comparisonMatrix)),ncol=ncol(comparisonMatrix),dimnames = dimnames(comparisonMatrix))
+    
+    if (!is.numeric(comparisonMatrix)) {
+      addComment("[ERROR]Filtering matrix is not fully numeric!",T,opt$log,display=FALSE)
+      q( "no", 1, F )
+    }
+    
+    if(expressionToCluster && length(setdiff(rownames(comparisonMatrix),rownames(expressionMatrix)))!=0){
+      addComment("[WARNING]All genes from filtering file are not included in expression file",T,opt$log,display=FALSE)
+    }
+    
+    if(expressionToCluster && length(setdiff(rownames(expressionMatrix),rownames(comparisonMatrix)))!=0){
+      addComment("[WARNING]All genes from expression file are not included in filtering file",T,opt$log,display=FALSE)
+    }
+    
+    addComment("[INFO]Filtering file loaded and checked",T,opt$log,display=FALSE)
+    addComment(c("[INFO]Dim of original filtering matrix:",dim(comparisonMatrix)),T,opt$log,display=FALSE)
+    
+    #restrict to user specified comparisons
+    restrictedComparisons=c()
+    if(!is.null(opt$comparisonNameLow))restrictedComparisons=unique(c(restrictedComparisons,unlist(strsplit(opt$comparisonNameLow,","))))
+    if(!is.null(opt$comparisonNameHigh))restrictedComparisons=unique(c(restrictedComparisons,unlist(strsplit(opt$comparisonNameHigh,","))))
+    
+    if (!all(restrictedComparisons%in%colnames(comparisonMatrix))){
+      addComment("[ERROR]Selected columns in filtering file are not present in filtering matrix!",T,opt$log,display=FALSE)
+      q( "no", 1, F )
+    }
+    comparisonMatrix=matrix(comparisonMatrix[,restrictedComparisons],ncol=length(restrictedComparisons),dimnames = list(rownames(comparisonMatrix),restrictedComparisons))
+    
+    #get number of required comparisons
+    nbComparisons=ncol(comparisonMatrix)
+    
+    addComment(c("[INFO]Dim of effective filtering matrix:",dim(comparisonMatrix)),T,opt$log,display=FALSE)
+}
+
+
+
+factorInfoMatrix=NULL
+if(!is.null(opt$factorInfo)){
+  #get group information
+  #load factors file
+  factorInfoMatrix=read.csv(file=opt$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]
+  
+  factorBarColor=colnames(factorInfoMatrix)[2]
+  
+  if(ncol(factorInfoMatrix)>2){
+    addComment("[ERROR]Factors file should not contain more than 2 columns",T,opt$log,display=FALSE)
+    q( "no", 1, F )
+  }
+  
+  #factor file is used for color band on heatmap, so all expression matrix column should be in the factor file
+  if(expressionToCluster && length(setdiff(colnames(expressionMatrix),rownames(factorInfoMatrix)))!=0){
+    addComment("[ERROR]Missing samples in factor file",T,opt$log,display=FALSE)
+    q( "no", 1, F )
+  }
+  
+  #factor file is used for color band on heatmap, so all comparison matrix column should be in the factor file
+  if(!expressionToCluster && length(setdiff(colnames(comparisonMatrix),rownames(factorInfoMatrix)))!=0){
+    addComment("[ERROR]Missing differential contrasts in factor file",T,opt$log,display=FALSE)
+    q( "no", 1, F )
+  }
+  
+  addComment("[INFO]Factors OK",T,opt$log,display=FALSE)
+  addComment(c("[INFO]Dim of factorInfo matrix:",dim(factorInfoMatrix)),T,opt$log,display=FALSE)
+}
+
+if(!is.null(opt$personalColors)){
+ ##parse personal colors
+  personalColors=unlist(strsplit(opt$personalColors,","))
+  if(length(personalColors)==2){
+    ##add medium color between two to get three colors
+    personalColors=c(personalColors[1],paste(c("#",as.character(as.hexmode(floor(apply(col2rgb(personalColors),1,mean))))),collapse=""),personalColors[2])
+  }
+  if(length(personalColors)!=3){
+    addComment("[ERROR]Personalized colors doesn't contain enough colors",T,opt$log,display=FALSE)
+    q( "no", 1, F )
+  }
+    
+}
+
+
+if(!is.null(opt$filterInputOutput) && opt$filterInputOutput=="input"){
+  #filter input data
+  
+    if(is.null(opt$geneListFiltering)){
+      #filtering using stat thresholds
+      #rowToKeep=intersect(which(comparisonMatrix[,seq(2,ncol(comparisonMatrix),4)]<=opt$pvalThreshold),which(abs(comparisonMatrix[,seq(4,ncol(comparisonMatrix),4)])>=log2(opt$FCthreshold)))
+      if(is.null(opt$genericData)){
+        #diff. expression matrix
+        rowToKeep=names(which(unlist(apply(comparisonMatrix,1,function(x)length(intersect(which(x[seq(2,length(x),nbColPerContrast)]<opt$pvalThreshold),which(abs(x[seq(4,length(x),nbColPerContrast)])>log2(opt$FCthreshold))))!=0))))
+      }else{
+        #generic filtering matrix
+        rowToKeep=rownames(comparisonMatrix)
+        if(!is.null(opt$comparisonNameLow)){
+          restrictedLowComparisons=unlist(strsplit(opt$comparisonNameLow,","))
+          rowToKeep=intersect(rowToKeep,names(which(unlist(apply(comparisonMatrix,1,function(x)length(which(x[restrictedLowComparisons]>opt$FCthreshold))!=0)))))
+        }
+        if(!is.null(opt$comparisonNameHigh)){
+          restrictedHighComparisons=unlist(strsplit(opt$comparisonNameHigh,","))
+          rowToKeep=intersect(rowToKeep,names(which(unlist(apply(comparisonMatrix,1,function(x)length(which(x[restrictedHighComparisons]<opt$pvalThreshold))!=0)))))
+        }
+      }
+    }else{
+      #filtering using user gene list
+      geneListFiltering=read.csv(opt$geneListFiltering,as.is = 1,header=F)
+      rowToKeep=unlist(c(geneListFiltering))
+    }
+    
+    if(!is.null(comparisonMatrix) && !all(rowToKeep%in%rownames(comparisonMatrix))){
+      #should arrive only with user gene list filtering with diff.exp. results clustering
+      addComment("[WARNING] some genes of the user defined list are not in the diff. exp. input file",T,opt$log)
+      rowToKeep=intersect(rowToKeep,rownames(comparisonMatrix))
+    }
+  
+    if(expressionToCluster && !all(rowToKeep%in%rownames(expressionMatrix))){
+      addComment("[WARNING] some genes selected by the input filter are not in the expression file",T,opt$log)
+      rowToKeep=intersect(rowToKeep,rownames(expressionMatrix))
+    }
+  
+    if(length(rowToKeep)==0){
+      addComment("[ERROR]No gene survived to the input filtering thresholds, execution will be aborted.
+                 Please consider to change threshold values and re-run the tool.",T,opt$log)
+      q( "no", 1, F )
+    }
+
+    #filter comparison matrix 
+    if(!is.null(comparisonMatrix)){
+      comparisonMatrix=matrix(comparisonMatrix[rowToKeep,],ncol=ncol(comparisonMatrix),dimnames = list(rowToKeep,colnames(comparisonMatrix)))
+      if(!is.null(comparisonMatrixInfoGene))comparisonMatrixInfoGene=comparisonMatrixInfoGene[rowToKeep]
+    }
+    #then expression matrix
+    if(expressionToCluster)expressionMatrix=matrix(expressionMatrix[rowToKeep,],ncol=ncol(expressionMatrix),dimnames = list(rowToKeep,colnames(expressionMatrix)))
+
+    if(!is.null(comparisonMatrix) && expressionToCluster && nrow(comparisonMatrix)!=nrow(expressionMatrix)){
+      addComment("[ERROR]Problem during input filtering, please check code",T,opt$log,display=FALSE)
+      q( "no", 1, F )
+    }
+    
+    addComment("[INFO]Filtering step done",T,opt$log,display=FALSE)
+    addComment(c("[INFO]Input filtering step:",length(rowToKeep),"remaining rows"),T,opt$log,display=FALSE)
+}
+
+
+addComment("[INFO]Ready to plot",T,opt$log,display=FALSE)
+
+##---------------------
+
+#plot heatmap
+if(expressionToCluster){
+  #will make clustering based on expression value or generic value
+  dataToHeatMap=expressionMatrix
+  valueMeaning="Intensity"
+  if(!is.null(opt$genericData))valueMeaning="Value"
+}else{
+  #will make clustering on log2(FC) values
+  dataToHeatMap=matrix(comparisonMatrix[,seq(4,ncol(comparisonMatrix),nbColPerContrast)],ncol=nbComparisons,dimnames = list(rownames(comparisonMatrix),colnames(comparisonMatrix)[seq(1,ncol(comparisonMatrix),nbColPerContrast)]))
+  valueMeaning="Log2(FC)"
+}
+addComment(c("[INFO]Dim of heatmap matrix:",dim(dataToHeatMap)),T,opt$log,display=FALSE)
+
+if(nrow(dataToHeatMap)==1 && ncol(dataToHeatMap)==1){
+  addComment("[ERROR]Cannot make clustering with unique cell tab",T,opt$log,display=FALSE)
+  q( "no", 1, F )
+}
+
+
+#apply data transformation if needed
+if(opt$dataTransformation=="log"){
+  dataToHeatMap=log(dataToHeatMap)
+  valueMeaning=paste(c("log(",valueMeaning,")"),collapse="")
+  addComment("[INFO]Data to cluster and to display in the heatmap are log transformed",T,opt$log,display=FALSE)
+}
+if(opt$dataTransformation=="log2"){
+  dataToHeatMap=log2(dataToHeatMap)
+  valueMeaning=paste(c("log2(",valueMeaning,")"),collapse="")
+  addComment("[INFO]Data to cluster and to display in the heatmap are log2 transformed",T,opt$log,display=FALSE)
+}
+
+maxRowsToDisplay=opt$maxRows
+
+nbClusters=opt$clusterNumber
+if(nbClusters>nrow(dataToHeatMap)){
+  #correct number of clusters if needed
+  nbClusters=nrow(dataToHeatMap)
+  addComment(c("[WARNING]Not enough rows to reach required clusters number, it is reduced to number of rows:",nbClusters),T,opt$log,display=FALSE)
+}
+
+nbSampleClusters=opt$sampleClusterNumber
+if(nbSampleClusters>ncol(dataToHeatMap)){
+  #correct number of clusters if needed
+  nbSampleClusters=ncol(dataToHeatMap)
+  addComment(c("[WARNING]Not enough columns to reach required conditions clusters number, it is reduced to number of columns:",nbSampleClusters),T,opt$log,display=FALSE)
+}
+
+colClust=FALSE
+rowClust=FALSE
+effectiveRowClust=FALSE
+
+#make appropriate clustering if needed
+if(nrow(dataToHeatMap)>1 && nbClusters>1)rowClust=hclust(distExtended(dataToHeatMap,method = opt$distanceMeasure),method = opt$aggloMethod)
+if(ncol(dataToHeatMap)>1 && nbSampleClusters>1)colClust=hclust(distExtended(t(dataToHeatMap),method = opt$distanceMeasure),method = opt$aggloMethod)
+
+if(nrow(dataToHeatMap)>maxRowsToDisplay){
+  #make subsampling based on preliminary global clustering
+  #clusteringResults=cutree(rowClust,nbClusters)
+  #heatMapGenesToKeep=unlist(lapply(seq(1,nbClusters),function(x)sample(which(clusteringResults==x),min(length(which(clusteringResults==x)),round(maxRowsToDisplay/nbClusters)))))
+  ##OR
+  #basic subsampling
+  heatMapGenesToKeep=sample(rownames(dataToHeatMap),maxRowsToDisplay)
+  effectiveDataToHeatMap=matrix(dataToHeatMap[heatMapGenesToKeep,],ncol=ncol(dataToHeatMap),dimnames=list(heatMapGenesToKeep,colnames(dataToHeatMap)))
+  effectiveNbClusters=min(nbClusters,maxRowsToDisplay)
+  if(nrow(effectiveDataToHeatMap)>1 && effectiveNbClusters>1)effectiveRowClust=hclust(distExtended(effectiveDataToHeatMap, method = opt$distanceMeasure),method = opt$aggloMethod)
+  addComment(c("[WARNING]Too many rows for efficient heatmap drawing",maxRowsToDisplay,"subsampling is done for vizualization only"),T,opt$log,display=FALSE)
+  rm(heatMapGenesToKeep)
+}else{
+  effectiveDataToHeatMap=dataToHeatMap
+  effectiveRowClust=rowClust 
+  effectiveNbClusters=nbClusters
+}
+
+addComment(c("[INFO]Dim of plotted heatmap matrix:",dim(effectiveDataToHeatMap)),T,opt$log,display=FALSE)
+
+personalized_hoverinfo=matrix("",ncol = ncol(effectiveDataToHeatMap),nrow = nrow(effectiveDataToHeatMap),dimnames = dimnames(effectiveDataToHeatMap))
+if(expressionToCluster){
+  for(iCol in colnames(effectiveDataToHeatMap)){for(iRow in rownames(effectiveDataToHeatMap)){personalized_hoverinfo[iRow,iCol]=paste(c("Probe: ",iRow,"\nCondition: ",iCol,"\n",valueMeaning,": ",effectiveDataToHeatMap[iRow,iCol]),collapse="")}}
+}else{
+  for(iCol in colnames(effectiveDataToHeatMap)){for(iRow in rownames(effectiveDataToHeatMap)){personalized_hoverinfo[iRow,iCol]=paste(c("Probe: ",iRow,"\nCondition: ",iCol,"\nFC: ",round(2^effectiveDataToHeatMap[iRow,iCol],2)),collapse="")}}
+}
+
+#trying to overcome limitation of heatmaply package to modify xtick and ytick label, using directly plotly functions, but for now plotly do not permit to have personalized color for each x/y tick separately
+test=FALSE
+if(test==TRUE){
+  
+  #define dendogram shapes
+  dd.row <- as.dendrogram(effectiveRowClust)
+  dd.col <- as.dendrogram(colClust)
+  
+  #and color them
+  dd.row=color_branches(dd.row, k = effectiveNbClusters, groupLabels = T)
+  dd.col=color_branches(dd.col, k = nbSampleClusters, groupLabels = T)
+  
+  #generating function for dendogram from segment list
+  ggdend <- function(df) {
+    ggplot() +
+      geom_segment(data = df, aes(x=x, y=y, xend=xend, yend=yend)) +
+      labs(x = "", y = "") + theme_minimal() +
+      theme(axis.text = element_blank(), axis.ticks = element_blank(),
+            panel.grid = element_blank())
+  }
+  
+  # generate x/y dendogram plots
+  px <- ggdend(dendro_data(dd.col)$segments)
+  py <- ggdend(dendro_data(dd.row)$segments) + coord_flip()
+  
+  # reshape data matrix
+  col.ord <- order.dendrogram(dd.col)
+  row.ord <- order.dendrogram(dd.row)
+  xx <- effectiveDataToHeatMap[row.ord, col.ord]
+  # and also personalized_hoverinfo
+  personalized_hoverinfo=personalized_hoverinfo[row.ord, col.ord]
+  
+  # hide axis ticks and grid lines
+  eaxis <- list(
+    showticklabels = FALSE,
+    showgrid = FALSE,
+    zeroline = FALSE
+  )
+  
+  #make the empty plot
+  p_empty <- plot_ly() %>%
+    layout(margin = list(l = 200),
+           xaxis = eaxis,
+           yaxis = eaxis)
+  
+  heatmap.plotly <- plot_ly(
+    z = xx, x = 1:ncol(xx), y = 1:nrow(xx), colors = viridis(n = 101, alpha = 1, begin = 0, end = 1, option = "inferno"),
+    type = "heatmap", showlegend = FALSE, text = personalized_hoverinfo, hoverinfo = "text",
+    colorbar = list(
+      # Capitalise first letter
+      title = valueMeaning,
+      tickmode = "array",
+      len = 0.3
+    )
+  ) %>%
+    layout(
+      xaxis = list(
+        tickfont = list(size = 10,color=get_leaves_branches_col(dd.row)),
+        tickangle = 45,
+        tickvals = 1:ncol(xx), ticktext = colnames(xx),
+        linecolor = "#ffffff",
+        range = c(0.5, ncol(xx) + 0.5),
+        showticklabels = TRUE
+      ),
+      yaxis = list(
+        tickfont = list(size = 10, color=get_leaves_branches_col(dd.col)),
+        tickangle = 0,
+        tickvals = 1:nrow(xx), ticktext = rownames(xx),
+        linecolor = "#ffffff",
+        range = c(0.5, nrow(xx) + 0.5),
+        showticklabels = TRUE
+      )
+    )
+  
+  #generate plotly 
+  pp <- subplot(px, p_empty, heatmap.plotly, py, nrows = 2, margin = 0,widths = c(0.8,0.2),heights = c(0.2,0.8), shareX = TRUE, 
+                shareY = TRUE)
+  
+  #save image file
+  export(pp, file =  paste(c(file.path(getwd(), "plotDir"),"/Heatmap.",opt$format),collapse=""))
+  #rise a bug due to token stuf
+  #orca(pp, file =  paste(c(file.path(getwd(), "plotDir"),"/Heatmap.",opt$format),collapse=""))
+  
+  
+  #save plotLy file
+  htmlwidgets::saveWidget(as_widget(pp), paste(c(file.path(getwd(), "plotLyDir"),"/Heatmap.html"),collapse=""),selfcontained = F)
+  
+  #htmlwidgets::saveWidget(as_widget(pp),"~/Bureau/test.html",selfcontained = F)
+  
+}else{ #test
+  label_names=c("Probe","Condition",valueMeaning)
+  
+  # #color hclust objects
+  # dd.row=color_branches(effectiveRowClust, k = effectiveNbClusters)
+  # #rowColors=get_leaves_branches_col(dd.row)
+  # #rowColors[order.dendrogram(dd.row)]=rowColors
+  # rowGroup=cutree(effectiveRowClust, k = effectiveNbClusters)
+  # 
+  # #get order of class as they will be displayed on the dendogram
+  # rowGroupRenamed=data.frame(cluster=mapvalues(rowGroup, unique(rowGroup[order.dendrogram(dd.row)[nleaves(dd.row):1]]), 1:effectiveNbClusters))
+  #
+  #  dd.col=color_branches(colClust, k = nbSampleClusters)
+  #  #colColors=get_leaves_branches_col(dd.col)
+  #  #colColors[order.dendrogram(dd.col)]=colColors
+  #  colGroup=cutree(colClust, k = nbSampleClusters)
+  #  
+  # # #get order of class as they will be displayed on the dendogram
+  #  colGroupRenamed=data.frame(sampleCluster=mapvalues(colGroup, unique(colGroup[order.dendrogram(dd.col)[nleaves(dd.col):1]]), 1:nbSampleClusters))
+  
+  
+  #while option is not correctly managed by heatmap apply, put personalized_hoverinfo to NULL
+  personalized_hoverinfo=NULL
+  
+  if(is.null(opt$personalColors)){
+    heatmapColors=viridis(n = 101, alpha = 1, begin = 0, end = 1, option = "inferno")
+  }else{
+    heatmapColors=personalColors
+  }
+  
+  colGroupRenamed=NULL
+  if(!is.null(factorInfoMatrix)){
+    colGroupRenamed=eval(parse(text=(paste("data.frame(",factorBarColor,"=factorInfoMatrix[colnames(effectiveDataToHeatMap),2])",sep=""))))
+    sideBarGroupNb=length(table(factorInfoMatrix[colnames(effectiveDataToHeatMap),2]))
+    sideBarColorPaletteName="Spectral"
+    if(!is.null(opt$sideBarColorPalette) && opt$sideBarColorPalette%in%rownames(RColorBrewer::brewer.pal.info)){
+      sideBarColorPaletteName=opt$sideBarColorPalette
+    }
+    sideBarColorPalette=setNames(colorRampPalette(RColorBrewer::brewer.pal(RColorBrewer::brewer.pal.info[sideBarColorPaletteName,"maxcolors"], sideBarColorPaletteName))(sideBarGroupNb),unique(factorInfoMatrix[colnames(effectiveDataToHeatMap),2]))
+  }
+  
+  if(!is.null(colGroupRenamed)){
+    pp <- heatmaply(effectiveDataToHeatMap,key.title = valueMeaning,k_row=effectiveNbClusters,k_col=nbSampleClusters,col_side_colors=colGroupRenamed,col_side_palette=sideBarColorPalette,Rowv=effectiveRowClust,Colv=colClust,label_names=label_names,custom_hovertext=personalized_hoverinfo,plot_method = "plotly",colors = heatmapColors)
+  }else{
+    pp <- heatmaply(effectiveDataToHeatMap,key.title = valueMeaning,k_row=effectiveNbClusters,k_col=nbSampleClusters,Rowv=effectiveRowClust,Colv=colClust,label_names=label_names,custom_hovertext=personalized_hoverinfo,plot_method = "plotly",colors = heatmapColors)
+  }
+  
+  
+  #save image file
+  export(pp, file =  paste(c(file.path(getwd(), "plotDir"),"/Heatmap.",opt$format),collapse=""))
+  #rise a bug due to token stuf
+  #orca(pp, file =  paste(c(file.path(getwd(), "plotDir"),"/Heatmap.",opt$format),collapse=""))
+  
+  
+  #save plotLy file
+  htmlwidgets::saveWidget(as_widget(pp), paste(c(file.path(getwd(), "plotLyDir"),"/Heatmap.html"),collapse=""),selfcontained = F)
+  
+}
+addComment("[INFO]Heatmap drawn",T,opt$log,display=FALSE)  
+
+
+#plot circular heatmap
+if(!class(effectiveRowClust)=="logical"){
+  dendo=as.dendrogram(effectiveRowClust)
+  
+  if(is.null(opt$personalColors)){
+    col_fun = colorRamp2(quantile(effectiveDataToHeatMap,probs = seq(0,1,0.01)), viridis(101,option = "inferno"))
+  }else{
+    col_fun = colorRamp2(quantile(effectiveDataToHeatMap,probs = seq(0,1,0.5)), personalColors)
+  }
+  
+  if(opt$format=="pdf"){
+    pdf(paste(c("./plotDir/circularPlot.pdf"),collapse=""))}else{
+      png(paste(c("./plotDir/circularPlot.png"),collapse=""))
+    }
+  
+  circos.par(cell.padding = c(0, 0, 0, 0), gap.degree = 5)
+  circos.initialize(c(rep("a",nrow(effectiveDataToHeatMap)),"b"),xlim=cbind(c(0,0),c(nrow(effectiveDataToHeatMap),5)))
+  circos.track(ylim = c(0, 1), bg.border = NA, panel.fun = function(x, y) {
+    if(CELL_META$sector.index=="a"){
+      nr = ncol(effectiveDataToHeatMap)
+      nc = nrow(effectiveDataToHeatMap)
+      circos.text(1:nc- 0.5, rep(0,nc), adj = c(0, 0), 
+                  rownames(effectiveDataToHeatMap)[order.dendrogram(dendo)], facing = "clockwise", niceFacing = TRUE, cex = 0.3)
+    }
+  })
+  
+  circos.track(ylim = c(0, ncol(effectiveDataToHeatMap)), bg.border = NA, panel.fun = function(x, y) {
+    
+    m = t(matrix(effectiveDataToHeatMap[order.dendrogram(dendo),],ncol=ncol(effectiveDataToHeatMap)))
+    col_mat = col_fun(m)
+    nr = nrow(m)
+    nc = ncol(m)
+    if(CELL_META$sector.index=="a"){
+      for(i in 1:nr) {
+        circos.rect(1:nc - 1, rep(nr - i, nc), 
+                    1:nc, rep(nr - i + 1, nc), 
+                    border = col_mat[i, ], col = col_mat[i, ])
+      }
+    }else{
+      circos.text(rep(1,nr), seq(nr,1,-1) , colnames(effectiveDataToHeatMap),cex = 0.3)
+    }
+  })
+  
+  #dendo = color_branches(dendo, k = effectiveNbClusters, col = colorRampPalette(brewer.pal(12,"Set3"))(effectiveNbClusters))
+  dendo = color_branches(dendo, k = effectiveNbClusters, col = rev(colorspace::rainbow_hcl(effectiveNbClusters)))
+  
+  
+  circos.track(ylim = c(0, attributes(dendo)$height), bg.border = NA, track.height = 0.25, 
+               panel.fun = function(x, y) {
+                 if(CELL_META$sector.index=="a")circos.dendrogram(dendo)} )
+  
+  circos.clear()
+  ##add legend
+  lgd_links = Legend(at = seq(ceiling(min(effectiveDataToHeatMap)),floor(max(effectiveDataToHeatMap)),ceiling((floor(max(effectiveDataToHeatMap))-ceiling(min(effectiveDataToHeatMap)))/4)), col_fun = col_fun, 
+                     title_position = "topleft", grid_width = unit(5, "mm") ,title = valueMeaning)
+  
+  pushViewport(viewport(x = 0.85, y = 0.80, 
+                        width = 0.1, 
+                        height = 0.1, 
+                        just = c("left", "bottom")))
+  grid.draw(lgd_links)
+  upViewport()
+  
+  
+  dev.off()
+  
+  addComment("[INFO]Circular heatmap drawn",T,opt$log,display=FALSE)  
+  loc <- Sys.setlocale("LC_NUMERIC","C")
+}else{
+  addComment(c("[WARNING]Circular plot will not be plotted considering row or cluster number < 2"),T,opt$log,display=FALSE)
+}
+rm(effectiveDataToHeatMap,effectiveRowClust,effectiveNbClusters)
+
+#plot screeplot 
+if(class(rowClust)!="logical" && nrow(dataToHeatMap)>2){
+  screePlotData=c()
+  for(iNbClusters in 2:(nbClusters+min(10,max(0,nrow(dataToHeatMap)-nbClusters)))){
+    clusteringResults=cutree(rowClust,iNbClusters)
+    #clusteringResults=kmeans(dataToHeatMap,iNbClusters)$cluster
+    
+    #compute variance between each intra-class points amongst themselves (need at least 3 points by cluster)
+    #screePlotData=c(screePlotData,sum(unlist(lapply(seq(1,iNbClusters),function(x){temp=which(clusteringResults==x);if(length(temp)>2){var(dist(dataToHeatMap[temp,]))}else{0}}))) )
+    #compute variance between each intra-class points and fictive mean point (need at least 2 points by cluster)
+    #screePlotData=c(screePlotData,sum(unlist(lapply(seq(1,iNbClusters),function(x){temp=which(clusteringResults==x);if(length(temp)>1){   var(dist(rbind(apply(dataToHeatMap[temp,],2,mean),dataToHeatMap[temp,]))[1:length(temp)]) }else{0}}))) )
+    if(ncol(dataToHeatMap)>1)screePlotData=c(screePlotData,sum(unlist(lapply(seq(1,iNbClusters),function(x){temp=which(clusteringResults==x);if(length(temp)>1){   sum((distExtended(rbind(apply(dataToHeatMap[temp,],2,mean),dataToHeatMap[temp,]),method = opt$distanceMeasure)[1:length(temp)])^2) }else{0}}))) )
+    else screePlotData=c(screePlotData,sum(unlist(lapply(seq(1,iNbClusters),function(x){temp=which(clusteringResults==x);if(length(temp)>1){   sum((dataToHeatMap[temp,]-mean(dataToHeatMap[temp,]))^2) }else{0}}))) )
+  }
+  
+  dataToPlot=data.frame(clusterNb=seq(2,length(screePlotData)+1),wcss=screePlotData)
+  p <- ggplot(data=dataToPlot, aes(clusterNb,wcss)) + geom_point(colour="#EE4444") + geom_line(colour="#DD9999") +
+    ggtitle("Scree plot") + theme_bw() + xlab(label="Cluster number") + ylab(label="Within cluster sum of squares") + 
+    theme(panel.border=element_blank(),plot.title = element_text(hjust = 0.5),legend.position = "none") +
+    scale_x_continuous(breaks=seq(min(dataToPlot$clusterNb), max(dataToPlot$clusterNb), 1))
+  
+  #save plotly files   
+  pp <- ggplotly(p)
+  
+  if(opt$format=="pdf"){
+    pdf(paste(c("./plotDir/screePlot.pdf"),collapse=""))}else{
+      png(paste(c("./plotDir/screePlot.png"),collapse=""))
+    }
+  plot(p)
+  dev.off()
+  
+  #save plotly files 
+  htmlwidgets::saveWidget(as_widget(pp), paste(c(file.path(getwd(), "plotLyDir"),"/screePlot.html"),collapse=""),selfcontained = F)
+  
+  addComment("[INFO]Scree plot drawn",T,opt$log,display=FALSE)  
+}else{
+  addComment(c("[WARNING]Scree plot will not be plotted considering row number <= 2"),T,opt$log,display=FALSE)
+}
+
+##----------------------
+  
+#filter output based on parameters
+
+rowToKeep=rownames(dataToHeatMap)
+if(!is.null(opt$filterInputOutput) && opt$filterInputOutput=="output"){
+  #rowToKeep=intersect(which(comparisonMatrix[,seq(2,ncol(comparisonMatrix),4)]<=opt$pvalThreshold),which(abs(comparisonMatrix[,seq(4,ncol(comparisonMatrix),4)])>=log2(opt$FCthreshold)))
+  if(is.null(opt$geneListFiltering)){
+    if(is.null(opt$genericData)){
+      #diff. expression matrix
+      rowToKeep=names(which(unlist(apply(comparisonMatrix,1,function(x)length(intersect(which(x[seq(2,length(x),nbColPerContrast)]<=opt$pvalThreshold),which(abs(x[seq(4,length(x),nbColPerContrast)])>=log2(opt$FCthreshold))))!=0))))
+    }else{
+      #generic filtering matrix
+      rowToKeep=rownames(comparisonMatrix)
+      if(!is.null(opt$comparisonNameLow)){
+        restrictedLowComparisons=unlist(strsplit(opt$comparisonNameLow,","))
+        rowToKeep=intersect(rowToKeep,names(which(unlist(apply(comparisonMatrix,1,function(x)length(which(x[restrictedLowComparisons]>opt$FCthreshold))!=0)))))
+      }
+      if(!is.null(opt$comparisonNameHigh)){
+        restrictedHighComparisons=unlist(strsplit(opt$comparisonNameHigh,","))
+        rowToKeep=intersect(rowToKeep,names(which(unlist(apply(comparisonMatrix,1,function(x)length(which(x[restrictedHighComparisons]<opt$pvalThreshold))!=0)))))
+      }
+    }
+  }else{
+    geneListFiltering=read.csv(opt$geneListFiltering,as.is = 1,header=F)
+    rowToKeep=unlist(c(geneListFiltering))
+  }
+  if(!is.null(comparisonMatrix) && !all(rowToKeep%in%rownames(comparisonMatrix))){
+    #should arrive only with user gene list filtering with diff.exp. results clustering
+    addComment("[WARNING] some genes of the user defined list are not in the diff. exp. input file",T,opt$log)
+    rowToKeep=intersect(rowToKeep,rownames(comparisonMatrix))
+  }
+  
+  if(expressionToCluster && !all(rowToKeep%in%rownames(expressionMatrix))){
+    addComment("[WARNING] some genes selected by the output filter are not in the expression file",T,opt$log)
+    rowToKeep=intersect(rowToKeep,rownames(expressionMatrix))
+  }
+  addComment(c("[INFO]Output filtering step:",length(rowToKeep),"remaining rows"),T,opt$log,display=FALSE) 
+}
+
+#we add differential analysis info in output if it was directly used for clustering or when it was used for filtering with expression
+
+#in case of expression or generic data clustering without filtering based on external stats
+if(expressionToCluster && is.null(comparisonMatrix)){
+  if(length(rowToKeep)==0){
+    addComment("[WARNING]No more gene after output filtering step, tabular output will be empty",T,opt$log,display=FALSE)
+    outputData=matrix(c("Gene","Cluster","noGene","noClustering"),ncol=2,nrow=2,byrow = TRUE)
+  }else{
+      outputData=matrix(0,ncol=2,nrow=length(rowToKeep)+1)
+      outputData[1,]=c("Gene","Cluster")
+      outputData[2:(length(rowToKeep)+1),1]=rowToKeep
+      if(class(rowClust)!="logical" ){
+        outputData[2:(length(rowToKeep)+1),2]=cutree(rowClust,nbClusters)[rowToKeep]
+      }else{
+        outputData[2:(length(rowToKeep)+1),2]=0
+      }
+  }
+}
+
+#in case of generic data clustering with filtering based on generic external data
+if(!is.null(opt$genericData) && !is.null(comparisonMatrix)){
+  if(length(rowToKeep)==0){
+    addComment("[WARNING]No more gene after output filtering step, tabular output will be empty",T,opt$log,display=FALSE)
+    outputData=matrix(c("Gene","Cluster","noGene","noClustering"),ncol=2,nrow=2,byrow = TRUE)
+  }else{
+    outputData=matrix(0,ncol=2+nbComparisons,nrow=length(rowToKeep)+1)
+    outputData[1,]=c("Gene","Cluster",colnames(comparisonMatrix))
+    outputData[2:(length(rowToKeep)+1),1]=rowToKeep
+    if(class(rowClust)!="logical" ){
+      outputData[2:(length(rowToKeep)+1),2]=cutree(rowClust,nbClusters)[rowToKeep]
+    }else{
+      outputData[2:(length(rowToKeep)+1),2]=0
+    }
+    outputData[2:(length(rowToKeep)+1),3:(ncol(comparisonMatrix)+2)]=prettyNum(comparisonMatrix[rowToKeep,],digits=4)
+  }
+}
+
+#in case of expression data clustering with filtering based on diff. exp. results or diff. exp. results clustering
+if(is.null(opt$genericData) && !is.null(comparisonMatrix)){
+  if(length(rowToKeep)==0){
+    addComment("[WARNING]No more gene after output filtering step, tabular output will be empty",T,opt$log,display=FALSE)
+    outputData=matrix(0,ncol=3,nrow=3)
+    outputData[1,]=c("","","Comparison")
+    outputData[2,]=c("Gene","Info","Cluster")
+    outputData[3,]=c("noGene","noInfo","noClustering")
+  }else{
+      outputData=matrix(0,ncol=3+nbComparisons*nbColPerContrast,nrow=length(rowToKeep)+2)
+      outputData[1,]=c("","","Comparison",rep(colnames(comparisonMatrix)[seq(1,ncol(comparisonMatrix),nbColPerContrast)],each=nbColPerContrast))
+      outputData[2,]=c("Gene","Info","Cluster",rep(c("p-val","FDR.p-val","FC","log2(FC)","t-stat"),nbComparisons))
+      outputData[3:(length(rowToKeep)+2),1]=rowToKeep
+      outputData[3:(length(rowToKeep)+2),2]=comparisonMatrixInfoGene[rowToKeep]
+      if(class(rowClust)!="logical" ){
+        outputData[3:(length(rowToKeep)+2),3]=cutree(rowClust,nbClusters)[rowToKeep]
+      }else{
+        outputData[3:(length(rowToKeep)+2),3]=0
+      }
+      outputData[3:(length(rowToKeep)+2),4:(ncol(comparisonMatrix)+3)]=prettyNum(comparisonMatrix[rowToKeep,],digits=4)
+  }
+}
+
+addComment("[INFO]Formated output",T,opt$log,display=FALSE) 
+write.table(outputData,file=opt$outputFile,quote=FALSE,sep="\t",col.names = F,row.names = F)
+  
+##----------------------
+
+end.time <- Sys.time()
+addComment(c("[INFO]Total execution time for R script:",as.numeric(end.time - start.time,units="mins"),"mins"),T,opt$log,display=FALSE)
+
+
+addComment("[INFO]End of R script",T,opt$log,display=FALSE)
+
+printSessionInfo(opt$log)
+
+#sessionInfo()
+
+
+