--- /dev/null
+GeneSimuAt<-function(DVDconstant=NULL, DVDqt1=NULL, DVDqt2=NULL, Conditions, NumofSample, NumofGene=NULL, DEGeneProp, Phiconstant=NULL, Phi.qt1=NULL, Phi.qt2=NULL, Meanconstant=NULL,NormFactor=NULL, OnlyData=T)
+{
+# 2012 feb 1
+# paired level simulation
+
+data(GeneEBresultGouldBart2)
+if(is.null(NormFactor)) NormFactor=rep(1,NumofSample)
+
+#MeansC1=rowMeans(GeneV.norm1.NZ.b2[,1:4])
+#MeansC2=rowMeans(GeneV.norm1.NZ.b2[,5:8])
+MeansC1=GeneEBresultGouldBart2$C1Mean[[1]]
+MeansC2=GeneEBresultGouldBart2$C2Mean[[1]]
+
+MeanDVD=MeansC1/MeansC2
+
+if(is.null(DVDconstant))DVDLibrary=MeanDVD[MeanDVD<quantile(MeanDVD[MeanDVD!=Inf],DVDqt2) & MeanDVD>quantile(MeanDVD[MeanDVD!=Inf],DVDqt1)]
+
+
+# If DVD constant, use constant when generate
+# If not, use DVDLibrary
+
+MeanInputraw=GeneEBresultGouldBart2$MeanList[[1]]
+#MeanInputraw=rowMeans(GeneV.norm1.NZ.b2)
+#Var1=apply(GeneV.norm1.NZ.b2[,1:4],1,var)
+#Var2=apply(GeneV.norm1.NZ.b2[,5:8],1,var)
+#VarInput=(Var1 + Var2)/2
+#If NumofGene.raw=NULL, empirical # of Gene
+#If !=NULL , Input a 9-vector
+NumofGene.raw=length(MeanInputraw)
+
+# here phi denotes r -- which is 1/phi' in which sigma^2=mu(1+mu phi')
+# In negative binomial
+# size is 1/phi'
+# rnbinom(100,size=100,mu=10)
+# var(qq)
+#[1] 10.93687
+# qq=rnbinom(100,size=10,mu=10)
+# var(qq)
+#[1] 24.01404
+
+#PhiInput.raw=(MeanInputraw^2) / (VarInput - MeanInputraw)
+PhiInput.raw=GeneEBresultGouldBart2$RList[[1]]
+if (length(Phiconstant)==0){
+ PhiLibrary=PhiInput.raw[1/(PhiInput.raw)<quantile(1/(PhiInput.raw),Phi.qt2) & 1/(PhiInput.raw)>quantile(1/(PhiInput.raw),Phi.qt1)]
+ PhiInputNames=sample(names(PhiLibrary),NumofGene.raw,replace=T)
+ PhiInput=PhiInput.raw[PhiInputNames]
+
+
+}
+
+if (length(Phiconstant)!=0)PhiInput=rep(Phiconstant,length(MeanInputraw))
+if(length(Meanconstant)==0)MeanInput=GeneEBresultGouldBart2$MeanList[[1]][PhiInputNames]
+if(length(Meanconstant)!=0)MeanInput=rep(Meanconstant,length(GeneEBresultGouldBart2$MeanList[[1]]))
+
+# Wanna DENumbers be proportion to 2
+DEGeneNumbers=round(NumofGene.raw*DEGeneProp/2)*2
+GeneNames=paste("G",c(1:NumofGene.raw),sep="_")
+names(PhiInput)=GeneNames
+names(MeanInput)=GeneNames
+
+#########
+# data
+#########
+EEList=sapply(1:NumofGene.raw, function(j) sapply(1:NumofSample, function(i)rnbinom(1,mu=NormFactor[i]*MeanInput[j], size=PhiInput[j])))
+
+
+
+
+ generateDataraw=t(EEList)
+ if(length(DVDconstant)==0){
+ DVDSample=sample(DVDLibrary,DEGeneNumbers,replace=T)
+ for(j in 1:NumofGene.raw){
+ if (j<=(DEGeneNumbers/2)) generateDataraw[j,((NumofSample/2)+1):NumofSample]=sapply(((NumofSample/2) +1):NumofSample, function(i)rnbinom(1, size=PhiInput[j], mu=DVDSample[j]*MeanInput[j]*NormFactor[i]),simplify=T)
+ if (j>=((DEGeneNumbers/2)+1) & j <=DEGeneNumbers) generateDataraw[j,1:(NumofSample/2)]=sapply(1:(NumofSample/2),function(i)rnbinom(1, size=MeanInput[j], mu= DVDSample[j]*MeanInput[j]*NormFactor[i]),simplify=T)
+}
+ }
+ if(length(DVDconstant)!=0){
+ for(j in 1:NumofGene.raw){
+ if (j<=(DEGeneNumbers/2)) generateDataraw[j,((NumofSample/2)+1):NumofSample]=sapply((NumofSample/2+1):NumofSample, function(i)rnbinom(1, size=MeanInput[j],mu=DVDconstant*MeanInput[j]*NormFactor[i]),simplify=T)
+ if (j>=((DEGeneNumbers/2)+1) & j <=DEGeneNumbers) generateDataraw[j,1:(NumofSample/2)]=sapply(1:(NumofSample/2),function(i)rnbinom(1, size=MeanInput[j],mu=DVDconstant*MeanInput[j]*NormFactor[i]),simplify=T)
+ }
+ }
+rownames(generateDataraw)=GeneNames
+MeanVector=rowMeans(generateDataraw)
+VarVector=apply(generateDataraw,1,var)
+MOV.post=MeanVector/VarVector
+
+
+
+### Remove MOV=NA
+generateData=generateDataraw
+generateData=generateData[!is.na(MOV.post)& MeanVector>2 & MeanVector<10000 ,]
+print(paste("NA MOV's",sum(is.na(MOV.post)),sum( MeanVector<2), sum(MeanVector>10000)))
+## DE
+NumDENow=sum(rownames(generateData)%in%rownames(generateDataraw)[1:DEGeneNumbers])
+
+if(length(NumofGene)!=0)
+ generateData=generateData[c(sample(1:NumDENow,round(NumofGene*DEGeneProp),replace=F),round( (dim(generateData)[1]+1-NumofGene*(1-DEGeneProp)):dim(generateData)[1])),]
+
+
+UseName=rownames(generateData)
+
+TrueDE=UseName[UseName%in%rownames(generateDataraw)[1:DEGeneNumbers]]
+phiuse=PhiInput[rownames(generateData)]
+meanuse=MeanInput[rownames(generateData)]
+
+#ArtiNames=rownames(generateData)[(DEGeneNumbers+1):(2*DEGeneNumbers)]
+#Noise=sample(c(1,ncol(generateData)),DEGeneNumbers,replace=T)
+TrueDELength=length(TrueDE)
+AtLoc=sample(c(1:length(Conditions)), TrueDELength, replace=T)
+AtFold=sample(c(4,6,8,10),TrueDELength, replace=T)
+
+AtNames_Level=vector("list",4)
+names(AtNames_Level)=c(4,6,8,10)
+for(i in 1:TrueDELength){
+generateData[(TrueDELength+i),AtLoc[i]]=generateData[(TrueDELength+i),AtLoc[i]]*AtFold[i]
+AtNames_Level[[as.character(AtFold[i])]]=c(AtNames_Level[[as.character(AtFold[i])]],rownames(generateData)[TrueDELength+i])
+}
+
+
+if(OnlyData==T){
+ OutName=paste("Gene",c(1:nrow(generateData)),sep="_")
+ names(OutName)=rownames(generateData)
+ OutData=generateData
+ rownames(OutData)=as.vector(OutName)
+ OutAt=as.vector(OutName[AtNames_Level])
+ OutTrueDE=as.vector(OutName[TrueDE])
+ output=list(data=OutData, TrueDE=OutTrueDE,Outliers=OutAt)
+ return(output)
+ }
+## DESeq
+
+cds=newCountDataSet(round(generateData),Conditions)
+cds=estimateSizeFactors(cds)
+Sizes=sizeFactors(cds)
+if(dim(generateData)[2]>4)cds=estimateVarianceFunctions(cds)
+else cds=estimateVarianceFunctions(cds, method="blind")
+
+res=nbinomTest(cds, "1", "2")
+ResAdj=res$padj
+names(ResAdj)=res$id
+SmallPValueName=names(ResAdj)[which(ResAdj<=.05)]
+print(paste("DESEq found",length(SmallPValueName)))
+print(paste("In True DE",sum(SmallPValueName%in%TrueDE)))
+
+print("DESeq Size factors")
+print(Sizes)
+
+## DESeq each group
+## Ours
+NewData=generateData
+
+
+#source("/z/Comp/kendziorskigroup/ningleng/RNASEQ/CODE/FinalV/NBBetaBiasUniqueP_PoolVar_SpeedUp_MDFPoi_NoNormVar.R")
+#source("/z/Comp/kendziorskigroup/ningleng/RNASEQ/CODE/FinalV/NBBetaBiasUniqueP_PoolVar_SpeedUp_MDFPoi_NoNormPoolR.R")
+
+EBresult=EBTest(NewData,rep(1,dim(NewData)[1]), rep(1,dim(NewData)[1]), rep(1,dim(NewData)[1]),Conditions,sizeFactors=Sizes,5)
+
+#EBres2=NBBetaEB.bias.uniqueP_PoolVarSpeedUp_MDFPoi_NoNormPoolR(NewData,rep(1,dim(NewData)[1]), rep(1,dim(NewData)[1]), rep(1,dim(NewData)[1]),Conditions,sizeFactors=Sizes,5)
+
+
+zlist.unlist=EBresult[[5]]
+fdr=max(.5,crit_fun(1-zlist.unlist,.05))
+EBDE=names(zlist.unlist)[which(zlist.unlist>fdr)]
+EBDE.Poi=names(EBresult[[6]])[which(EBresult[[6]]>fdr)]
+zlist.unlist.whole=c(EBresult[[5]],EBresult[[6]])
+print(paste("Soft EB Poi",length(EBDE.Poi)))
+EBDE=c(EBDE, EBDE.Poi)
+print(paste("Soft EB found",length(EBDE)))
+print(paste("In True DE",sum(EBDE%in%TrueDE)))
+
+EBDE95=names(zlist.unlist)[which(zlist.unlist>.95)]
+EBDE95.Poi=names(EBresult[[6]])[which(EBresult[[6]]>.95)]
+print(paste("Hard Poi found",length(EBDE95.Poi)))
+EBDE95=c(EBDE95, EBDE95.Poi)
+print(paste("Hard EB found" ,length(EBDE95)))
+print(paste("In True DE",sum(EBDE95%in%TrueDE)))
+
+### edgeR
+library(edgeR,lib.loc="~/RCODE")
+edgeRList.b2=DGEList(NewData,group=Conditions)
+if(length(Phiconstant)==1){
+ edgeRList.b2=estimateCommonDisp(edgeRList.b2)
+ edgeRRes.b2=exactTest(edgeRList.b2)
+}
+if(length(Phiconstant)==0){
+ edgeRList.b2=estimateCommonDisp(edgeRList.b2)
+ edgeRList.b2=estimateTagwiseDisp(edgeRList.b2)
+ edgeRRes.b2=exactTest(edgeRList.b2, common.disp = FALSE)
+}
+edgeRPvalue.b2.raw=edgeRRes.b2[[1]][[3]]
+edgeRPvalue.b2=p.adjust(edgeRPvalue.b2.raw, method="BH")
+names(edgeRPvalue.b2)=rownames(NewData)
+edgeRSmallpvalue=names(which(edgeRPvalue.b2<.05))
+print(paste("edgeR found",length(edgeRSmallpvalue)))
+print(paste("In True DE",sum(edgeRSmallpvalue%in%TrueDE)))
+
+### Bayseq
+library(baySeq, lib.loc="~/RCODE")
+library(snow, lib.loc="~/RCODE")
+cl <- makeCluster(4, "SOCK")
+groups <- list(NDE = rep(1,NumofSample), DE = rep(c(1,2),each=NumofSample/2))
+CD <- new("countData", data = NewData, replicates = Conditions, libsizes = as.integer(colSums(NewData)), groups = groups)
+CDP.NBML <- getPriors.NB(CD, samplesize = dim(NewData)[1], estimation = "QL", cl = cl)
+CDPost.NBML <- getLikelihoods.NB(CDP.NBML, pET = "BIC", cl = cl)
+bayseqPost=CDPost.NBML@posteriors
+rownames(bayseqPost)=rownames(NewData)
+bayseqDE=rownames(NewData)[bayseqPost[,2]>log(.95)]
+print(paste("bayseq found",length(bayseqDE)))
+print(paste("In True DE",sum(bayseqDE%in%TrueDE)))
+
+
+### BBSeq
+library("BBSeq",lib.loc="~/RCODE")
+CondM=cbind(rep(1,NumofSample),rep(c(0,1),each=NumofSample/2))
+output=free.estimate(NewData,CondM)
+beta.free = output$betahat.free
+p.free = output$p.free
+psi.free = output$psi.free
+names(p.free)=rownames(NewData)
+p.free.adj=p.adjust(p.free,method="BH")
+# Top p free?
+#out.model=constrained.estimate(NewData,CondM, gn=3, beta.free ,psi.free)
+#p.constrained = out.model$p.model
+BBDE=names(p.free.adj)[which(p.free.adj<.05)]
+print(paste("BBSeq found",length(BBDE)))
+print(paste("In True DE",sum(BBDE%in%TrueDE)))
+
+
+#########################
+# Generate table
+Table=matrix(rep(0,12),ncol=2)
+colnames(Table)=c("Power","FDR")
+rownames(Table)=c("DESeq","edgeR","BaySeq","BBSeq","EBSeq_ModifiedSoft","EBSeq_Hard")
+
+ Length=length(TrueDE)
+ Table[1,1]=sum(SmallPValueName%in%TrueDE)/Length
+ Table[2,1]=sum(edgeRSmallpvalue%in%TrueDE)/Length
+ Table[3,1]=sum(bayseqDE%in%TrueDE)/Length
+ Table[4,1]=sum(BBDE%in%TrueDE)/Length
+ Table[5,1]=sum(EBDE%in%TrueDE)/Length
+ Table[6,1]=sum(EBDE95%in%TrueDE)/Length
+ Table[1,2]=sum(!SmallPValueName%in%TrueDE)/length(SmallPValueName)
+ Table[2,2]=sum(!edgeRSmallpvalue%in%TrueDE)/length(edgeRSmallpvalue)
+ Table[3,2]=sum(!bayseqDE%in%TrueDE)/length(bayseqDE)
+ Table[4,2]=sum(!BBDE%in%TrueDE)/length(BBDE)
+ Table[5,2]=sum(!EBDE%in%TrueDE)/length(EBDE)
+ Table[6,2]=sum(!EBDE95%in%TrueDE)/length(EBDE95)
+ Table=round(Table,2)
+
+ValueTable=matrix(rep(0,12),ncol=2)
+colnames(ValueTable)=c("Power","FDR")
+rownames(ValueTable)=c("DESeq","edgeR","BaySeq","BBSeq","EBSeq_ModifiedSoft","EBSeq_Hard")
+ ValueTable[1,1]=sum(SmallPValueName%in%TrueDE)
+ ValueTable[2,1]=sum(edgeRSmallpvalue%in%TrueDE)
+ ValueTable[3,1]=sum(bayseqDE%in%TrueDE)
+ ValueTable[4,1]=sum(BBDE%in%TrueDE)
+ ValueTable[5,1]=sum(EBDE%in%TrueDE)
+ ValueTable[6,1]=sum(EBDE95%in%TrueDE)
+ ValueTable[1,2]=sum(!SmallPValueName%in%TrueDE)
+ ValueTable[2,2]=sum(!edgeRSmallpvalue%in%TrueDE)
+ ValueTable[3,2]=sum(!bayseqDE%in%TrueDE)
+ ValueTable[4,2]=sum(!BBDE%in%TrueDE)
+ ValueTable[5,2]=sum(!EBDE%in%TrueDE)
+ ValueTable[6,2]=sum(!EBDE95%in%TrueDE)
+
+
+AtFoundTable=matrix(rep(0,24),ncol=4)
+colnames(AtFoundTable)=paste("Level",c(1:4),sep="_")
+rownames(Table)=c("DESeq","edgeR","BaySeq","BBSeq","EBSeq_ModifiedSoft","EBSeq_Hard")
+for(i in 1:4){
+ AtFoundTable[1,i]=sum(SmallPValueName%in%AtNames_Level[[i]])
+ AtFoundTable[2,i]=sum(edgeRSmallpvalue%in%AtNames_Level[[i]])
+ AtFoundTable[3,i]=sum(bayseqDE%in%AtNames_Level[[i]])
+ AtFoundTable[4,i]=sum(BBDE%in%AtNames_Level[[i]])
+ AtFoundTable[5,i]=sum(EBDE%in%AtNames_Level[[i]])
+ AtFoundTable[6,i]=sum(EBDE95%in%AtNames_Level[[i]])
+ }
+
+
+if(length(DVDconstant)==0)DVD=c(quantile(MeanDVD[MeanDVD!=Inf],DVDqt1), quantile(MeanDVD[MeanDVD!=Inf],DVDqt2))
+if(length(DVDconstant)!=0) DVD=DVDconstant
+if(length(Phiconstant)==0)Phi=c(quantile(PhiInput.raw,Phi.qt1), quantile(PhiInput.raw,Phi.qt2))
+if(length(Phiconstant)!=0) Phi=Phiconstant
+OUT=list(Table=Table, ValueTable=ValueTable, DVD=DVD, Phi=Phi, generateData=NewData, TrueDE=TrueDE,phi.vector=phiuse,mean.vector=meanuse,NormFactor=NormFactor, DESeqP=ResAdj, edgeRP=edgeRPvalue.b2, EBSeqPP=zlist.unlist.whole, BaySeqPP=bayseqPost,BBSeqP=p.free.adj,EBoutput=EBresult, AtFoundTable= AtFoundTable,Outliers=AtNames_Level)
+
+
+
+}
+
+
projects / rsem.git / blobdiff