Nothing
R/EBHMMNBMultiEM_2chain.R
In EBSeqHMM: Bayesian analysis for identifying gene or isoform expression changes in ordered RNA-seq experiments
Defines functions
EBHMMNBMultiEM_2chain
Documented in
EBHMMNBMultiEM_2chain
#' @title Run EBSeqHMM model with a fixed expected FC
#' @usage EBHMMNBMultiEM_2chain(Data,
#' NgVector=NULL, Conditions, AllTran=NULL,
#' AllPi0=NULL, Terms=NULL,
#' sizeFactors, NumTranStage=c(3,2),PriorFC=2,
#' StateNames=c("Up","Down"),homo=FALSE,
#' UpdateRd=5, PIBound=.9, UpdatePI=FALSE,Print=FALSE,
#' WithinCondR=TRUE,
#' PenalizeLowMed=TRUE, PenalizeLowMedQt=.1,PenalizeLowMedVal=10)
#' @param Data input data, rows are genes and columns are samples
#' @param NgVector Ng vector; NULL for gene level data
#' @param Conditions A factor indicates the condition (time/spatial point) which each sample belongs to.
#' @param AllTran initial values for transition matrices
#' @param AllPi0 initial values for starting probabilities
#' @param Terms Terms
#' @param sizeFactors a vector indicates library size factors
#' @param StateNames names of the hidden states
#' @param NumTranStage number of states in two chains
#' @param PriorFC target FC for gridient change
#' @param homo whether the chain is assumed to be homogenious
#' @param UpdateRd max number of iteration
#' @param UpdatePI whether update the mixture proportion of two chains
#' @param PIBound upper bound of the mixture proportion of the two chains
#' @param Print Whether print the elapsed-time while running the test.
#' @param WithinCondR By defining WithinCondR=TRUE, estimation of r's are obtained within each condition. (WithinCondR=FALSE is not suggested here)
#' @param PenalizeLowMed,PenalizeLowMedQt,PenalizeLowMedVal
#' Transcripts with median quantile < = PenalizeLowMedQt will be penalized
#' @author Ning Leng
#' @examples
#' data(GeneExampleData)
#' CondVector <- rep(paste("t",1:5,sep=""),each=3)
#' Conditions <- factor(CondVector, levels=c("t1","t2","t3","t4","t5"))
#' Sizes <- MedianNorm(GeneExampleData)
#' tmp <- EBHMMNBMultiEM_2chain(Data=GeneExampleData,sizeFactors=Sizes, Conditions=Conditions,
#' UpdateRd=2)
#' @details EBHMMNBMultiEM_2chain() function implements the EBSeqHMM model to perform statistical analysis in an RNA-seq experiment with ordered conditions.
#' EBHMMNBMultiEM_2chain() calls EBHMMNBfunForMulti() function to perform Balm-Welch algorithm that estimates the starting probabilities and transition probabilities.
#' Here the emission distribution of each gene is assumed to be a
#' Beta-Negative Binomial distribution with parameters (r_g, alpha, beta)
#' , in which alpha and beta are shared by all the genes and r_g is gene specific.
#' If not specified, r_g, alpha and beta will be estimated using method of moments.
#' For isoform data, we assume isoforms from the same Ig group share the same beta^Ig. alpha is shared by all the isoforms and r_gi is isoform specific.
#' The user also needs to specify an expected FC.
#' Function EBSeqHMMTest() runs several models with varying FCs and returns the model with maximum likelihood.
#' @return
#' Pi0Out: estimated starting probabilities of each iteration.
#'
#' TranOut: estimated transition probabilities of each iteration.
#'
#' Pi: estimated probability of being each chain.
#'
#' Alpha, Beta: estimated alpha and beta(s).
#'
#' LLSum: log likelihood of the model.
#'
#' QList: estimated q's.
#'
#' MgAllPP: marginal PP for all paths.
#'
#' MgAllMAPChar: Most likely path based on MgAllPP.
#'
#' MgAllMaxVal: Highest PP based on MgAllPP.
#'
#' PPMatW: Posterior probabilities of being each of the chains.
#'
EBHMMNBMultiEM_2chain <- function(Data,
NgVector=NULL, Conditions,
AllTran=NULL,
AllPi0=NULL, Terms=NULL,
sizeFactors, NumTranStage=c(3,2),PriorFC=2,
StateNames=c("Up","Down"),homo=FALSE,
UpdateRd=5, PIBound=.9, UpdatePI=FALSE,Print=FALSE
,WithinCondR=TRUE,
PenalizeLowMed=TRUE, PenalizeLowMedQt=.1,PenalizeLowMedVal=10){
##########################
# Define transition parameters if they are not defined in inputs
# First chain is noise chain - 3 states Up EE Down
# Second chain is signal chain - 2 states Up Down
##########################
NumCond <- nlevels(as.factor(Conditions))
if(is.null(AllTran)){
AllTran <- vector("list",2)
AllTran[[1]] <- array(1/3,dim=c(3,3,NumCond-2))
AllTran[[2]] <- array(1/2,dim=c(2,2,NumCond-2))
}
if(is.null(AllPi0)){
AllPi0 <- vector("list",2)
AllPi0[[1]] <- rep(1/3,3)
AllPi0[[2]] <- rep(1/2,2)
}
##########################
# Initialization and removing non expressed genes
##########################
PI <- rep(1/(length(AllTran)), length(AllTran))
PI.up <- PI
PI.track <- PI
Tran.Up <- AllTran
Pi0.Up <- AllPi0
DataRename <- Data
NamesIn <- rownames(Data)
NewNames <- paste("g",1:nrow(Data),sep="")
rownames(DataRename) <- NewNames
Data.NZ <- DataRename[which(rowSums(Data)>0),]
Data.NZ.OriName <- Data[which(rowSums(Data)>0),]
if(!is.null(NgVector))names(NgVector) <- NewNames
NotNA <- c(1:nrow(Data.NZ))
StateNameList <- vector("list",2)
StateNameList[[1]] <- c("Up","EE","Down")
StateNameList[[2]] <- c("Up","Down")
UpParam <- c(FALSE,TRUE)
if(UpdateRd==0){
UpParam <- c(FALSE, FALSE)
UpdateRd <- 1
}
######################
# MOM alpha beta
######################
QList <- suppressWarnings(EBHMMNBfun(DataRename,NgVector=NgVector,Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=AllPi0[[1]], Tran=AllTran[[1]], NumTranStage=NumTranStage[1],
FCParam=NULL, AlphaIn=NULL,BetaIn=NULL,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=FALSE, Print=Print, OnlyQ=TRUE,WithinCondR=WithinCondR,
PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
QUse <- vector("list",length(QList)+1)
for(i in 1:length(QList))QUse[[i]] <- QList[[i]]$QList1
QUse[[length(QUse)]] <- QList[[length(QList)]]$QList2
# list length N, N=#conditions
QEach <- vector("list",length(QUse[[1]]))
for(i in 1:length(QUse[[1]]))
QEach[[i]] <- as.vector(unlist(sapply(QUse,function(j)j[[i]])))
QInRange <- sapply(QEach,function(i)i[which(i>0 & i<1)],simplify=FALSE)
Est <- sapply(QInRange,beta.mom,simplify=FALSE)
AlphaMom <- mean(unlist(sapply(Est,function(i)i[1])))
BetaMom <- unlist(sapply(Est,function(i)i[2]))
# Q for check
NumGp <- length(QUse[[1]])
QOut <- vector("list",NumGp)
for(i in 1:NumGp){
QOut[[i]] <- vector("list",length(QUse))
for(j in 1:length(QUse)) QOut[[i]][[j]] <- QUse[[j]][[i]]
}
##########################
# Iterations
##########################
for(rd in 1:UpdateRd){
EachEB = EachLike=vector("list",length(AllTran))
for(i in 1:length(AllTran)){
# First run - P(Noise group) = P(Signal group)
if(rd==1){
EachEB[[i]] <- suppressWarnings(EBHMMNBfun(DataRename,NgVector=NgVector,Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=AllPi0[[i]], Tran=AllTran[[i]], NumTranStage=NumTranStage[i],
FCParam=NULL, AlphaIn=AlphaMom,BetaIn=BetaMom,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=FALSE, Print=Print,
WithinCondR=WithinCondR,
PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
}
# other runs - P(N group) and P(S group) estimated from last step
if(rd>1){
# Isoform level
if(!is.null(NgVector))
EachEB[[i]] <- suppressWarnings(EBHMMNBfunForMulti(DataRename[NotNA,],PPtmp0[,i],
NgVector=NgVector[NotNA],Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=Pi0.Up[[i]], Tran=Tran.Up[[i]], NumTranStage=NumTranStage[i],
FCParam=NULL, AlphaIn=AlphaMom,BetaIn=BetaMom,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=UpParam[i],Print=Print,
WithinCondR=WithinCondR,PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
# gene level
if(is.null(NgVector))
EachEB[[i]] <- suppressWarnings(EBHMMNBfunForMulti(DataRename[NotNA,],PPtmp0[,i],
NgVector=NULL,Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=Pi0.Up[[i]], Tran=Tran.Up[[i]], NumTranStage=NumTranStage[i],
FCParam=NULL, AlphaIn=AlphaMom,BetaIn=BetaMom,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=UpParam[i],Print=Print,
WithinCondR=WithinCondR,
PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
}
}
# Update parameters
if(rd>1){
Tran.Up <- sapply(1:length(EachEB),function(i)EachEB[[i]]$TranTrack[[length(EachEB[[i]]$TranTrack)]], simplify=FALSE)
if(Print==TRUE)print(Tran.Up)
Pi0.Up <- sapply(1:length(EachEB),function(i)EachEB[[i]]$Pi0Track[length(EachEB[[i]]$TranTrack),], simplify=FALSE)
}
# Calculate likelihood
AllTerms <- EachEB[[1]][[3]]
# Allf P(path)*P(data|path)
EachLike64 <- sapply(1:length(AllTran),function(i)EachEB[[i]]$Allf,
simplify=FALSE)
EachSum64Mat <- matrix(0, ncol=length(AllTran), nrow=nrow(EachEB[[i]]$Allf))
rownames(EachSum64Mat) <-
rownames(EachEB[[i]]$Allf)
NotNA <- rownames(EachEB[[i]]$Allf)
for(i in 1:(length(AllTran)))EachSum64Mat[rownames(EachLike64[[i]]),i] <- rowSums(exp(EachLike64[[i]]))
LogSum64Mat <- log(EachSum64Mat)
# PI=rep(1/(length(AllTran)), length(AllTran))
# PI.up=PI
# PI.track=PI
# for(j in 1:maxround){
LogPI <- log(PI.up)
LogPILike <- t(t(LogSum64Mat)+LogPI)
PILike <- exp(LogPILike)
SumPILike <- rowSums(PILike)
PPtmp <- PILike/SumPILike
PI.up <- colMeans(PPtmp,na.rm=TRUE)
if(!is.null(PIBound)){
if(max(PI.up)>PIBound){
PI.up[which.max(PI.up)] <- PIBound
PI.up[which.min(PI.up)] <- 1-PIBound
}
}
if(UpdatePI==FALSE)PI.up <- PI
PI.track <- rbind(PI.track, PI.up)
# }
# Calc PP based on PI
LogPI <- log(PI.up)
LogPILike <- t(t(LogSum64Mat)+LogPI)
# likelihood
PILike <- exp(LogPILike)
SumPILike <- rowSums(PILike)
#################
# P (W=1 | Y)
##################
# PP
PPtmp0 <- PILike/SumPILike
# Max term
PPMaxT0 <- sapply(1:nrow(PPtmp0),function(i){
tt=which.max(PPtmp0[i,])
if(length(tt)==0)tt=NA
tt
})
PPMaxVal0 <- sapply(1:nrow(PPtmp0),function(i)max(PPtmp0[i,]))
}
#################
# P (Z=1 | Y)
##################
PostLikeLog_BG <- EachLike64[[1]]
PostLike_BG <- exp(PostLikeLog_BG) # P(path)P(Data|Path) in background chain
PostLikeMerge <- PostLike_BG*PI.up[1] # P(backgroud)P(Data|Background, path)
if(ncol(PostLikeMerge)==length(StateNameList[[1]]))colnames(PostLikeMerge) <- StateNameList[[1]]
# assume P(Data|path,signal) for EE related pathes are 0
for(i in colnames(EachLike64[[2]]))PostLikeMerge[,i] <- PostLikeMerge[,i]+PI.up[2]*exp(EachLike64[[2]][,i])
PPMerge <- PostLikeMerge/rowSums(PostLikeMerge)
WhichMax <- apply(PPMerge,1,function(oo){
tt=which.max(oo)
if(length(tt)>0)out=tt
if(length(tt)==0)out=NA
out})
WhichMax.NonNA.ind <- which(!is.na(WhichMax))
WhichMax.NonNA <- WhichMax[WhichMax.NonNA.ind]
MapTerm <- colnames(PPMerge)[WhichMax.NonNA]
MapVal <- sapply(1:length(WhichMax.NonNA),function(i)PPMerge[i,WhichMax.NonNA[i]])
names(MapTerm) = names(MapVal)=rownames(Data.NZ.OriName)[WhichMax.NonNA.ind]
#########################
# Rename
#########################
rownames(PPtmp0) = names(PPMaxT0)=names(PPMaxVal0)=
rownames(LogPILike)=rownames(PILike)=names(SumPILike)=
rownames(PPMerge)<-
rownames(Data.NZ.OriName)
# log likwlihood
# -Inf are imputed with smallest readable valyes
SumPILikeForLL <- log(SumPILike)
ISNA <- which(SumPILikeForLL==-Inf | is.na(SumPILikeForLL))
ISNOTNA <- c(1:length(SumPILikeForLL))[-ISNA]
Min.ISNOT <- min(SumPILikeForLL[ISNOTNA])
SumPILikeForLL[ISNA] <- Min.ISNOT
SumPILikeForLL.Sum <- sum(SumPILikeForLL)
####################
#Tran
####################
TranOut <- sapply(1:length(EachEB),function(i)EachEB[[i]]$TranTrack[[length(EachEB[[i]]$TranTrack)]],simplify=FALSE)
Pi0Out <- sapply(1:length(EachEB),function(i)EachEB[[i]]$Pi0Track[length(EachEB[[i]]$TranTrack),],simplify=FALSE)
####################
#
####################
# conditional probabilities
CondC1PPAll <- EachEB[[1]]$PP
CondC1PatAll <- EachEB[[1]]$AllTermChar
CondC1MAPChar <- apply(CondC1PPAll,1,function(i){
tt <- CondC1PatAll[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
CondC1MaxVal <- sapply(1:nrow(CondC1PPAll),function(i){
if(is.na(CondC1MAPChar[i]))tt <- NA
else tt <- CondC1PPAll[i,CondC1MAPChar[i]]
tt})
names(CondC1MaxVal)=rownames(CondC1PPAll)=
names(CondC1MAPChar)=rownames(Data.NZ.OriName)
CondC2PPAll <- EachEB[[2]]$PP
CondC2PatAll <- EachEB[[2]]$AllTermChar
CondC2MAPChar <- apply(CondC2PPAll,1,function(i){
tt <- CondC2PatAll[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
CondC2MaxVal <- sapply(1:nrow(CondC2PPAll),function(i){
if(is.na(CondC2MAPChar[i]))tt <- NA
else tt <- CondC2PPAll[i,CondC2MAPChar[i]]
tt})
names(CondC2MaxVal)=rownames(CondC2PPAll)=
names(CondC2MAPChar)=rownames(Data.NZ.OriName)
# Marginal probability of gradient patterns
PI <- PI.track
SigPPTerm <- colnames(CondC2PPAll)
SigPPCB <- (CondC2PPAll*PI[2]+CondC1PPAll[,SigPPTerm]*PI[1])
SigMAPCBChar <- apply(SigPPCB,1,function(i){
tt <- SigPPTerm[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
SigMaxValCB <- sapply(1:nrow(SigPPCB),function(i){
if(is.na(SigMAPCBChar[i]))tt <- NA
else tt <- SigPPCB[i,SigMAPCBChar[i]]
tt})
names(SigMAPCBChar)=names(SigMaxValCB)=rownames(Data.NZ.OriName)
# Marginal probability for all patterns
SigPPLarge <- cbind(SigPPCB,CondC1PPAll[,colnames(CondC1PPAll)[which(!colnames(CondC1PPAll)%in%colnames(SigPPCB))]]*PI[1])
SigMAPLargeChar <- apply(SigPPLarge,1,function(i){
tt <- colnames(SigPPLarge)[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
SigMaxValLarge <- sapply(1:nrow(SigPPLarge),function(i){
if(is.na(SigMAPLargeChar[i]))tt <- NA
else tt <- SigPPLarge[i,SigMAPLargeChar[i]]
tt})
names(SigMAPLargeChar)=names(SigMaxValLarge)=rownames(Data.NZ.OriName)
out <- list(
PPMatW=PPtmp0, PPMatZ=PPMerge,MAPZ=MapTerm, MAPVal=MapVal,
SignalPattern=colnames(EachLike64[[2]]),
LL=SumPILike,
PPMaxIdxW=PPMaxT0, PPMaxValW=PPMaxVal0, fW=PILike,
AllLike=EachLike64, PI=PI.track, TranOut=TranOut, Pi0Out=Pi0Out,
ResInChain=EachEB,LLEach=SumPILikeForLL,LLSum=SumPILikeForLL.Sum,
CondC1PP=CondC1PPAll, CondC1Pat=CondC1PatAll, CondC1MAPChar=CondC1MAPChar,
CondC1MaxVal=CondC1MaxVal,
CondC2PP=CondC2PPAll, CondC2Pat=CondC2PatAll, CondC2MAPChar=CondC2MAPChar,
CondC2MaxVal=CondC2MaxVal,
MgGrdPP=SigPPCB,MgGrdTerm=SigPPTerm, MgGrdMAPChar=SigMAPCBChar,
MgGrdMaxVal=SigMaxValCB, MgAllPP=SigPPLarge, MgAllMAPChar=SigMAPLargeChar,
MgAllMaxVal=SigMaxValLarge, QList=QOut, Alpha=matrix(AlphaMom,nrow=1),
Beta=matrix(BetaMom,nrow=1), PPpattern="PPpattern"
)
}
Try the EBSeqHMM package in your browser
Any scripts or data that you put into this service are public.
EBSeqHMM documentation built on Nov. 8, 2020, 5:22 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions EBHMMNBMultiEM_2chain
Documented in EBHMMNBMultiEM_2chain
#' @title Run EBSeqHMM model with a fixed expected FC
#' @usage EBHMMNBMultiEM_2chain(Data,
#' NgVector=NULL, Conditions, AllTran=NULL,
#' AllPi0=NULL, Terms=NULL,
#' sizeFactors, NumTranStage=c(3,2),PriorFC=2,
#' StateNames=c("Up","Down"),homo=FALSE,
#' UpdateRd=5, PIBound=.9, UpdatePI=FALSE,Print=FALSE,
#' WithinCondR=TRUE,
#' PenalizeLowMed=TRUE, PenalizeLowMedQt=.1,PenalizeLowMedVal=10)
#' @param Data input data, rows are genes and columns are samples
#' @param NgVector Ng vector; NULL for gene level data
#' @param Conditions A factor indicates the condition (time/spatial point) which each sample belongs to.
#' @param AllTran initial values for transition matrices
#' @param AllPi0 initial values for starting probabilities
#' @param Terms Terms
#' @param sizeFactors a vector indicates library size factors
#' @param StateNames names of the hidden states
#' @param NumTranStage number of states in two chains
#' @param PriorFC target FC for gridient change
#' @param homo whether the chain is assumed to be homogenious
#' @param UpdateRd max number of iteration
#' @param UpdatePI whether update the mixture proportion of two chains
#' @param PIBound upper bound of the mixture proportion of the two chains
#' @param Print Whether print the elapsed-time while running the test.
#' @param WithinCondR By defining WithinCondR=TRUE, estimation of r's are obtained within each condition. (WithinCondR=FALSE is not suggested here)
#' @param PenalizeLowMed,PenalizeLowMedQt,PenalizeLowMedVal
#' Transcripts with median quantile < = PenalizeLowMedQt will be penalized
#' @author Ning Leng
#' @examples
#' data(GeneExampleData)
#' CondVector <- rep(paste("t",1:5,sep=""),each=3)
#' Conditions <- factor(CondVector, levels=c("t1","t2","t3","t4","t5"))
#' Sizes <- MedianNorm(GeneExampleData)
#' tmp <- EBHMMNBMultiEM_2chain(Data=GeneExampleData,sizeFactors=Sizes, Conditions=Conditions,
#' UpdateRd=2)
#' @details EBHMMNBMultiEM_2chain() function implements the EBSeqHMM model to perform statistical analysis in an RNA-seq experiment with ordered conditions.
#' EBHMMNBMultiEM_2chain() calls EBHMMNBfunForMulti() function to perform Balm-Welch algorithm that estimates the starting probabilities and transition probabilities.
#' Here the emission distribution of each gene is assumed to be a
#' Beta-Negative Binomial distribution with parameters (r_g, alpha, beta)
#' , in which alpha and beta are shared by all the genes and r_g is gene specific.
#' If not specified, r_g, alpha and beta will be estimated using method of moments.
#' For isoform data, we assume isoforms from the same Ig group share the same beta^Ig. alpha is shared by all the isoforms and r_gi is isoform specific.
#' The user also needs to specify an expected FC.
#' Function EBSeqHMMTest() runs several models with varying FCs and returns the model with maximum likelihood.
#' @return
#' Pi0Out: estimated starting probabilities of each iteration.
#'
#' TranOut: estimated transition probabilities of each iteration.
#'
#' Pi: estimated probability of being each chain.
#'
#' Alpha, Beta: estimated alpha and beta(s).
#'
#' LLSum: log likelihood of the model.
#'
#' QList: estimated q's.
#'
#' MgAllPP: marginal PP for all paths.
#'
#' MgAllMAPChar: Most likely path based on MgAllPP.
#'
#' MgAllMaxVal: Highest PP based on MgAllPP.
#'
#' PPMatW: Posterior probabilities of being each of the chains.
#'
EBHMMNBMultiEM_2chain <- function(Data,
NgVector=NULL, Conditions,
AllTran=NULL,
AllPi0=NULL, Terms=NULL,
sizeFactors, NumTranStage=c(3,2),PriorFC=2,
StateNames=c("Up","Down"),homo=FALSE,
UpdateRd=5, PIBound=.9, UpdatePI=FALSE,Print=FALSE
,WithinCondR=TRUE,
PenalizeLowMed=TRUE, PenalizeLowMedQt=.1,PenalizeLowMedVal=10){
##########################
# Define transition parameters if they are not defined in inputs
# First chain is noise chain - 3 states Up EE Down
# Second chain is signal chain - 2 states Up Down
##########################
NumCond <- nlevels(as.factor(Conditions))
if(is.null(AllTran)){
AllTran <- vector("list",2)
AllTran[[1]] <- array(1/3,dim=c(3,3,NumCond-2))
AllTran[[2]] <- array(1/2,dim=c(2,2,NumCond-2))
}
if(is.null(AllPi0)){
AllPi0 <- vector("list",2)
AllPi0[[1]] <- rep(1/3,3)
AllPi0[[2]] <- rep(1/2,2)
}
##########################
# Initialization and removing non expressed genes
##########################
PI <- rep(1/(length(AllTran)), length(AllTran))
PI.up <- PI
PI.track <- PI
Tran.Up <- AllTran
Pi0.Up <- AllPi0
DataRename <- Data
NamesIn <- rownames(Data)
NewNames <- paste("g",1:nrow(Data),sep="")
rownames(DataRename) <- NewNames
Data.NZ <- DataRename[which(rowSums(Data)>0),]
Data.NZ.OriName <- Data[which(rowSums(Data)>0),]
if(!is.null(NgVector))names(NgVector) <- NewNames
NotNA <- c(1:nrow(Data.NZ))
StateNameList <- vector("list",2)
StateNameList[[1]] <- c("Up","EE","Down")
StateNameList[[2]] <- c("Up","Down")
UpParam <- c(FALSE,TRUE)
if(UpdateRd==0){
UpParam <- c(FALSE, FALSE)
UpdateRd <- 1
}
######################
# MOM alpha beta
######################
QList <- suppressWarnings(EBHMMNBfun(DataRename,NgVector=NgVector,Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=AllPi0[[1]], Tran=AllTran[[1]], NumTranStage=NumTranStage[1],
FCParam=NULL, AlphaIn=NULL,BetaIn=NULL,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=FALSE, Print=Print, OnlyQ=TRUE,WithinCondR=WithinCondR,
PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
QUse <- vector("list",length(QList)+1)
for(i in 1:length(QList))QUse[[i]] <- QList[[i]]$QList1
QUse[[length(QUse)]] <- QList[[length(QList)]]$QList2
# list length N, N=#conditions
QEach <- vector("list",length(QUse[[1]]))
for(i in 1:length(QUse[[1]]))
QEach[[i]] <- as.vector(unlist(sapply(QUse,function(j)j[[i]])))
QInRange <- sapply(QEach,function(i)i[which(i>0 & i<1)],simplify=FALSE)
Est <- sapply(QInRange,beta.mom,simplify=FALSE)
AlphaMom <- mean(unlist(sapply(Est,function(i)i[1])))
BetaMom <- unlist(sapply(Est,function(i)i[2]))
# Q for check
NumGp <- length(QUse[[1]])
QOut <- vector("list",NumGp)
for(i in 1:NumGp){
QOut[[i]] <- vector("list",length(QUse))
for(j in 1:length(QUse)) QOut[[i]][[j]] <- QUse[[j]][[i]]
}
##########################
# Iterations
##########################
for(rd in 1:UpdateRd){
EachEB = EachLike=vector("list",length(AllTran))
for(i in 1:length(AllTran)){
# First run - P(Noise group) = P(Signal group)
if(rd==1){
EachEB[[i]] <- suppressWarnings(EBHMMNBfun(DataRename,NgVector=NgVector,Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=AllPi0[[i]], Tran=AllTran[[i]], NumTranStage=NumTranStage[i],
FCParam=NULL, AlphaIn=AlphaMom,BetaIn=BetaMom,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=FALSE, Print=Print,
WithinCondR=WithinCondR,
PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
}
# other runs - P(N group) and P(S group) estimated from last step
if(rd>1){
# Isoform level
if(!is.null(NgVector))
EachEB[[i]] <- suppressWarnings(EBHMMNBfunForMulti(DataRename[NotNA,],PPtmp0[,i],
NgVector=NgVector[NotNA],Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=Pi0.Up[[i]], Tran=Tran.Up[[i]], NumTranStage=NumTranStage[i],
FCParam=NULL, AlphaIn=AlphaMom,BetaIn=BetaMom,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=UpParam[i],Print=Print,
WithinCondR=WithinCondR,PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
# gene level
if(is.null(NgVector))
EachEB[[i]] <- suppressWarnings(EBHMMNBfunForMulti(DataRename[NotNA,],PPtmp0[,i],
NgVector=NULL,Conditions=Conditions, sizeFactors=sizeFactors,
PriorFC=PriorFC,homo=homo, maxround=1,
Pi0=Pi0.Up[[i]], Tran=Tran.Up[[i]], NumTranStage=NumTranStage[i],
FCParam=NULL, AlphaIn=AlphaMom,BetaIn=BetaMom,
StateNames=StateNameList[[i]],
EM=FALSE, UpdateParam=UpParam[i],Print=Print,
WithinCondR=WithinCondR,
PenalizeLowMed=PenalizeLowMed,
PenalizeLowMedQt=PenalizeLowMedQt,
PenalizeLowMedVal=PenalizeLowMedVal))
}
}
# Update parameters
if(rd>1){
Tran.Up <- sapply(1:length(EachEB),function(i)EachEB[[i]]$TranTrack[[length(EachEB[[i]]$TranTrack)]], simplify=FALSE)
if(Print==TRUE)print(Tran.Up)
Pi0.Up <- sapply(1:length(EachEB),function(i)EachEB[[i]]$Pi0Track[length(EachEB[[i]]$TranTrack),], simplify=FALSE)
}
# Calculate likelihood
AllTerms <- EachEB[[1]][[3]]
# Allf P(path)*P(data|path)
EachLike64 <- sapply(1:length(AllTran),function(i)EachEB[[i]]$Allf,
simplify=FALSE)
EachSum64Mat <- matrix(0, ncol=length(AllTran), nrow=nrow(EachEB[[i]]$Allf))
rownames(EachSum64Mat) <-
rownames(EachEB[[i]]$Allf)
NotNA <- rownames(EachEB[[i]]$Allf)
for(i in 1:(length(AllTran)))EachSum64Mat[rownames(EachLike64[[i]]),i] <- rowSums(exp(EachLike64[[i]]))
LogSum64Mat <- log(EachSum64Mat)
# PI=rep(1/(length(AllTran)), length(AllTran))
# PI.up=PI
# PI.track=PI
# for(j in 1:maxround){
LogPI <- log(PI.up)
LogPILike <- t(t(LogSum64Mat)+LogPI)
PILike <- exp(LogPILike)
SumPILike <- rowSums(PILike)
PPtmp <- PILike/SumPILike
PI.up <- colMeans(PPtmp,na.rm=TRUE)
if(!is.null(PIBound)){
if(max(PI.up)>PIBound){
PI.up[which.max(PI.up)] <- PIBound
PI.up[which.min(PI.up)] <- 1-PIBound
}
}
if(UpdatePI==FALSE)PI.up <- PI
PI.track <- rbind(PI.track, PI.up)
# }
# Calc PP based on PI
LogPI <- log(PI.up)
LogPILike <- t(t(LogSum64Mat)+LogPI)
# likelihood
PILike <- exp(LogPILike)
SumPILike <- rowSums(PILike)
#################
# P (W=1 | Y)
##################
# PP
PPtmp0 <- PILike/SumPILike
# Max term
PPMaxT0 <- sapply(1:nrow(PPtmp0),function(i){
tt=which.max(PPtmp0[i,])
if(length(tt)==0)tt=NA
tt
})
PPMaxVal0 <- sapply(1:nrow(PPtmp0),function(i)max(PPtmp0[i,]))
}
#################
# P (Z=1 | Y)
##################
PostLikeLog_BG <- EachLike64[[1]]
PostLike_BG <- exp(PostLikeLog_BG) # P(path)P(Data|Path) in background chain
PostLikeMerge <- PostLike_BG*PI.up[1] # P(backgroud)P(Data|Background, path)
if(ncol(PostLikeMerge)==length(StateNameList[[1]]))colnames(PostLikeMerge) <- StateNameList[[1]]
# assume P(Data|path,signal) for EE related pathes are 0
for(i in colnames(EachLike64[[2]]))PostLikeMerge[,i] <- PostLikeMerge[,i]+PI.up[2]*exp(EachLike64[[2]][,i])
PPMerge <- PostLikeMerge/rowSums(PostLikeMerge)
WhichMax <- apply(PPMerge,1,function(oo){
tt=which.max(oo)
if(length(tt)>0)out=tt
if(length(tt)==0)out=NA
out})
WhichMax.NonNA.ind <- which(!is.na(WhichMax))
WhichMax.NonNA <- WhichMax[WhichMax.NonNA.ind]
MapTerm <- colnames(PPMerge)[WhichMax.NonNA]
MapVal <- sapply(1:length(WhichMax.NonNA),function(i)PPMerge[i,WhichMax.NonNA[i]])
names(MapTerm) = names(MapVal)=rownames(Data.NZ.OriName)[WhichMax.NonNA.ind]
#########################
# Rename
#########################
rownames(PPtmp0) = names(PPMaxT0)=names(PPMaxVal0)=
rownames(LogPILike)=rownames(PILike)=names(SumPILike)=
rownames(PPMerge)<-
rownames(Data.NZ.OriName)
# log likwlihood
# -Inf are imputed with smallest readable valyes
SumPILikeForLL <- log(SumPILike)
ISNA <- which(SumPILikeForLL==-Inf | is.na(SumPILikeForLL))
ISNOTNA <- c(1:length(SumPILikeForLL))[-ISNA]
Min.ISNOT <- min(SumPILikeForLL[ISNOTNA])
SumPILikeForLL[ISNA] <- Min.ISNOT
SumPILikeForLL.Sum <- sum(SumPILikeForLL)
####################
#Tran
####################
TranOut <- sapply(1:length(EachEB),function(i)EachEB[[i]]$TranTrack[[length(EachEB[[i]]$TranTrack)]],simplify=FALSE)
Pi0Out <- sapply(1:length(EachEB),function(i)EachEB[[i]]$Pi0Track[length(EachEB[[i]]$TranTrack),],simplify=FALSE)
####################
#
####################
# conditional probabilities
CondC1PPAll <- EachEB[[1]]$PP
CondC1PatAll <- EachEB[[1]]$AllTermChar
CondC1MAPChar <- apply(CondC1PPAll,1,function(i){
tt <- CondC1PatAll[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
CondC1MaxVal <- sapply(1:nrow(CondC1PPAll),function(i){
if(is.na(CondC1MAPChar[i]))tt <- NA
else tt <- CondC1PPAll[i,CondC1MAPChar[i]]
tt})
names(CondC1MaxVal)=rownames(CondC1PPAll)=
names(CondC1MAPChar)=rownames(Data.NZ.OriName)
CondC2PPAll <- EachEB[[2]]$PP
CondC2PatAll <- EachEB[[2]]$AllTermChar
CondC2MAPChar <- apply(CondC2PPAll,1,function(i){
tt <- CondC2PatAll[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
CondC2MaxVal <- sapply(1:nrow(CondC2PPAll),function(i){
if(is.na(CondC2MAPChar[i]))tt <- NA
else tt <- CondC2PPAll[i,CondC2MAPChar[i]]
tt})
names(CondC2MaxVal)=rownames(CondC2PPAll)=
names(CondC2MAPChar)=rownames(Data.NZ.OriName)
# Marginal probability of gradient patterns
PI <- PI.track
SigPPTerm <- colnames(CondC2PPAll)
SigPPCB <- (CondC2PPAll*PI[2]+CondC1PPAll[,SigPPTerm]*PI[1])
SigMAPCBChar <- apply(SigPPCB,1,function(i){
tt <- SigPPTerm[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
SigMaxValCB <- sapply(1:nrow(SigPPCB),function(i){
if(is.na(SigMAPCBChar[i]))tt <- NA
else tt <- SigPPCB[i,SigMAPCBChar[i]]
tt})
names(SigMAPCBChar)=names(SigMaxValCB)=rownames(Data.NZ.OriName)
# Marginal probability for all patterns
SigPPLarge <- cbind(SigPPCB,CondC1PPAll[,colnames(CondC1PPAll)[which(!colnames(CondC1PPAll)%in%colnames(SigPPCB))]]*PI[1])
SigMAPLargeChar <- apply(SigPPLarge,1,function(i){
tt <- colnames(SigPPLarge)[which.max(i)]
if(length(tt)==0)tt <- NA
tt})
SigMaxValLarge <- sapply(1:nrow(SigPPLarge),function(i){
if(is.na(SigMAPLargeChar[i]))tt <- NA
else tt <- SigPPLarge[i,SigMAPLargeChar[i]]
tt})
names(SigMAPLargeChar)=names(SigMaxValLarge)=rownames(Data.NZ.OriName)
out <- list(
PPMatW=PPtmp0, PPMatZ=PPMerge,MAPZ=MapTerm, MAPVal=MapVal,
SignalPattern=colnames(EachLike64[[2]]),
LL=SumPILike,
PPMaxIdxW=PPMaxT0, PPMaxValW=PPMaxVal0, fW=PILike,
AllLike=EachLike64, PI=PI.track, TranOut=TranOut, Pi0Out=Pi0Out,
ResInChain=EachEB,LLEach=SumPILikeForLL,LLSum=SumPILikeForLL.Sum,
CondC1PP=CondC1PPAll, CondC1Pat=CondC1PatAll, CondC1MAPChar=CondC1MAPChar,
CondC1MaxVal=CondC1MaxVal,
CondC2PP=CondC2PPAll, CondC2Pat=CondC2PatAll, CondC2MAPChar=CondC2MAPChar,
CondC2MaxVal=CondC2MaxVal,
MgGrdPP=SigPPCB,MgGrdTerm=SigPPTerm, MgGrdMAPChar=SigMAPCBChar,
MgGrdMaxVal=SigMaxValCB, MgAllPP=SigPPLarge, MgAllMAPChar=SigMAPLargeChar,
MgAllMaxVal=SigMaxValLarge, QList=QOut, Alpha=matrix(AlphaMom,nrow=1),
Beta=matrix(BetaMom,nrow=1), PPpattern="PPpattern"
)
}
Try the EBSeqHMM package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.