Mercurial > repos > fabio > iwtomics
view testandplot.R @ 18:565118df598a draft
Uploaded 20170531
author | fabio |
---|---|
date | Wed, 31 May 2017 11:07:22 -0400 |
parents | 1e677d6b1aaf |
children | 2bb6b44093ba |
line wrap: on
line source
if (require("IWTomics",character.only = TRUE,quietly = FALSE)) { args=commandArgs(TRUE) # get args names and values args_values=strsplit(args,'=') args_names=unlist(lapply(args_values,function(arg) arg[1])) names(args_values)=args_names args_values=lapply(args_values,function(arg) arg[2]) # read filenames adjustedpvaluematrix=args_values$adjustedpvaluematrix iwtomicsrespdf=args_values$iwtomicsrespdf iwtomicssumpdf=args_values$iwtomicssumpdf regionids=args_values$regionids featureids=args_values$featureids rdatafile=args_values$rdatafile iwtomicsrdata=args_values$iwtomicsrdata iwtomicstests=args_values$iwtomicstests iwtomicsselectedfeatures=args_values$iwtomicsselectedfeatures # read parameters (from region1 on) i_region1=which(args_names=='region1') for(i in i_region1:length(args)){ eval(parse(text=args[[i]])) } # load RData load(rdatafile) # read regionids and featureids regionids=as.character(read.delim(regionids,header=FALSE,sep='\t',stringsAsFactors=FALSE)) featureids=as.character(read.delim(featureids,header=FALSE,sep='\t',stringsAsFactors=FALSE)) # retrieve region1, region2 and features_subset ids and check they are in the RData id_region1=regionids[region1] id_region2=regionids[region2] id_features_subset=featureids[features_subset] if(length(setdiff(c(id_region1,id_region2),idRegions(regionsFeatures)))!=0){ quit(save="no", status=10) stop('Wrong region ids') } if(length(setdiff(id_features_subset,idFeatures(regionsFeatures)))!=0){ quit(save="no", status=20) stop('Wrong feature ids') } if(sum(duplicated(paste0(id_region1,id_region2)))){ quit(save="no", status=30) stop('Same test repeated multiple times.') } # perform test tryCatch({ # fix repeated probs if(statistics=='quantile'){ # fix repeated probs testprobs=sort(unique(testprobs)) }else{ testprobs=0.5 } regionsFeatures_test=IWTomicsTest(regionsFeatures,id_region1,id_region2,id_features_subset, statistics=statistics,probs=testprobs,B=B) # create adjustedvaluematrix output for(test in seq_along(id_region1)){ for(id_feature in id_features_subset){ write(paste0('Test: ',id_region1[test],' vs ',id_region2[test],', on feature ',id_feature), file=adjustedpvaluematrix,append=TRUE,sep='\t') pval=regionsFeatures_test@test$result[[test]][[id_feature]]$adjusted_pval_matrix row.names(pval)=paste('Scale',rev(seq_len(nrow(pval)))) write.table(pval,file=adjustedpvaluematrix,append=TRUE,sep='\t',quote=FALSE,row.names=TRUE,col.names=FALSE) write('',file=adjustedpvaluematrix,append=TRUE,sep='\t') } } }, error = function(err) { quit(save="no", status=40) #error testing stop(err) }) # plot test results pdf(iwtomicsrespdf,width=5,height=7) if(plottype=='boxplot'){ # fix repeated probs probs=sort(unique(probs)) }else{ probs=c(0.25,0.5,0.75) } plotTest(regionsFeatures_test,alpha=testalpha,type=plottype,probs=probs,average=average,size=size,ask=FALSE) dev.off() # plot summary results if(groupby!='none'){ pdf(iwtomicssumpdf,width=15,height=10) plotSummary(regionsFeatures_test,alpha=summaryalpha,only_significant=only_significant,groupby=groupby,ask=FALSE,append=TRUE) dev.off() } # create output write.table(as.data.frame(t(paste(id_region1,'vs',id_region2))),file=iwtomicstests,quote=FALSE,sep='\t',row.names=FALSE,col.names=FALSE) write.table(as.data.frame(t(idFeatures(regionsFeatures_test))),file=iwtomicsselectedfeatures,quote=FALSE,sep='\t',row.names=FALSE,col.names=FALSE) save(regionsFeatures_test,file=iwtomicsrdata) }else{ quit(save="no", status=255) stop("Missing IWTomics package") }