R/dpeakRead.R

In dpeak: dPeak (Deconvolution of Peaks in ChIP-seq Analysis)

# read bin-level data & process it

dpeakRead <- function( peakfile=NULL, readfile=NULL,
    fileFormat="eland_result", PET=FALSE, fragLen=200,
    parallel=FALSE, nCore=1, tempDir=NULL, perl="perl" )
{
    # process aligned read file

    if ( PET ) {
        message( "Info: Paired-end tag (PET) is assumed (PET=TRUE)." )
    } else {
        message( "Info: Single-end tag (SET) is assumed (PET=FALSE)." )
        message( "Info: Average fragment length is set as ",fragLen," (fragLen=",fragLen,")." )
    }

    message( "Info: Reading and processing aligned read file..." )

    if ( is.null(tempDir) ) {
        tempfileName <- tempfile( c("output","summary") )
    } else {
        tempfileName <- c( paste(tempDir,"output.txt",sep=""),
            paste(tempDir,"summary.txt",sep="") )
    }

        # intermediate file name
    .constructExtRead( infile=readfile, outfile=tempfileName[1],
        summaryfile=tempfileName[2], fileFormat=fileFormat,
        PET=PET, fragLen=fragLen, perl=perl )

    # read summary file (chrID, # lines)

    summaryInfo <- read.table( tempfileName[2], header=FALSE,
        stringsAsFactors=FALSE, comment.char="", check.names=FALSE )
    colnames(summaryInfo) <- c("chrID","nline")
    readChr <- summaryInfo[,1]

    # process peak set & reads (assume BED file format)
    # - peak set: assume that first 3 columns of both files are chr, start, end

    message( "Info: Reading peak list..." )
    peakSet <- read.table( peakfile, header=FALSE, stringsAsFactors=FALSE )
    nPeak <- nrow(peakSet)

    # match chromosomes between peak list and reads

    peakByChr <- split( peakSet[ , c(2,3), drop=FALSE ], peakSet[,1] )
    peakChr <- names(peakByChr)

    chrCommon <- Reduce( intersect, list( peakChr, readChr ) )
    chrCommon <- sort(chrCommon)

    # match reads to each peak region
    # (using parallel computing, if parallel exists)

    message( "Info: Processing and combining peak list and reads..." )

    if ( parallel == TRUE ) {
        out <- parallel::mclapply( chrCommon,
            function(x) .matchPeakRead( chr=x, peakCur=peakByChr[[x]],
                outfileName=paste(tempfileName[1],"_",x,sep=""),
                nRow=summaryInfo[ summaryInfo[,1]==x, 2 ], PET=PET ),
            mc.cores = nCore )
    } else {
        out <- lapply( chrCommon,
            function(x) .matchPeakRead( chr=x, peakCur=peakByChr[[x]],
                outfileName=paste(tempfileName[1],"_",x,sep=""),
                nRow=summaryInfo[ summaryInfo[,1]==x, 2 ], PET=PET )
            )
    }

    fragSet <- vector( "list", nPeak )
    peakChr <- peakStart <- peakEnd <- rep( NA, nPeak )
    nameVec <- rep( NA, nPeak )

    nEmpty <- 0
    emptyList <- c()
    cur <- 1

    if ( PET == FALSE ) {
        nF <- nAll <- rep( NA, length(chrCommon) )
    }

    for ( chr in seq_len(length(chrCommon)) ) {
        nPeakCur <- length( out[[chr]]$nameVecCur )

        for ( j in seq_len(nPeakCur) ) {
            if ( !is.na(out[[chr]]$fragSetCur[[j]][1,1]) ) {
                fragSet[[cur]] <- out[[chr]]$fragSetCur[[j]]
            } else {
                fragSet[[cur]] <- matrix( NA )
                nEmpty <- nEmpty + 1
                emptyList <- c( emptyList, out[[chr]]$nameVecCur[j] )
            }
            peakChr[cur] <- out[[chr]]$peakChrCur[j]
            peakStart[cur] <- out[[chr]]$peakStartCur[j]
            peakEnd[cur] <- out[[chr]]$peakEndCur[j]
            nameVec[cur] <- out[[chr]]$nameVecCur[j]
            cur <- cur + 1
        }

        if ( PET == FALSE ) {
            nF[chr] <- out[[chr]]$nFCur
            nAll[chr] <- out[[chr]]$nAllCur
        }
    }
    names(fragSet) <- nameVec

    if ( PET == TRUE ) {
        fragLen <- c()
        for ( chr in seq_len(length(chrCommon)) ) {
            fragLen <- c( fragLen, out[[chr]]$fragLenCur )
        }
    }

    if ( nEmpty == 0 ) {
        emptyList <- ""
    }

    # check proportion of forward reads for SET data

    if ( PET == TRUE ) {
        Fratio <- 0.5
    } else {
        Fratio <- sum(nF) / sum(nAll)
    }

    # stack fragment (projection to coordinates)

    if ( parallel == TRUE ) {
        stackedFragment <- parallel::mclapply( fragSet, .stackFragment, mc.cores = nCore )
    } else {
        stackedFragment <- lapply( fragSet, .stackFragment )
    }
    names(stackedFragment) <- nameVec

    # if PET, calculate distribution of fragment length

    if ( PET ) {
        message( "Info: Calculating distribution of fragment length..." )

        fragLenTable <- table( fragLen )
        aveFragLen <- median( fragLen )
    } else {
        fragLenTable <- table( fragLen )
        aveFragLen <- fragLen
    }

    message( "Info: Done.\n" )

    # remove temporary files after use

    unlink( tempfileName[1] )
    unlink( tempfileName[2] )

    # info about preprocessing

    nFrag <- unlist( lapply( fragSet,
        function(x) ifelse( !is.null(x), nrow(x), 0 )
     ) )
    sumRead <- sum(nFrag)
    medNumRead <- median(nFrag)

    message( "------------------------------------------------------------\n" )
    message( "Info: Preprocessing summary\n" )
    message( "------------------------------------------------------------\n" )
    message( "Tag type: ",ifelse(PET,"PET","SET"),"\n", sep="" )
    message( "Number of chromosomes: ",length(chrCommon),"\n", sep="" )
    message( "Number of peaks: ",nPeak,"\n", sep="" )
    message( "Number of peaks without reads: ",nEmpty,"\n", sep="" )
    message( "[Note] Use 'printEmpty' method to check the list.\n" )
    message( "Number of utilized reads: ",sumRead,"\n", sep="" )
    message( "Median number of reads in each peak: ",medNumRead,"\n", sep="" )
    if ( PET == TRUE ) {
        message( "Median fragment length: ",aveFragLen,"\n", sep="" )
    } else {
        message( "Fragment length (provided by user): ",aveFragLen,"\n", sep="" )
        Fper <- round( 100 * Fratio )
        Rper <- 100 - Fper
        message( "Percentage of forward strand reads: ",Fper," %\n", sep="" )
        message( "Percentage of reverse strand reads: ",Rper," %\n", sep="" )
    }
    message( "------------------------------------------------------------\n" )

    new( "DpeakData", fragSet=fragSet, PET=PET, fragLenTable=fragLenTable,
        aveFragLen=aveFragLen, Fratio=Fratio, stackedFragment=stackedFragment,
        peakChr=peakChr, peakStart=peakStart, peakEnd=peakEnd,
        emptyList=emptyList )
}

# match peak & reads

.matchPeakRead <- function( chr, peakCur, outfileName, nRow, PET ) {

    # read processed read file
    # - PET read: chr, start, end
    # - SET read: chr, position, strand, read length

    if ( PET == TRUE ) {
        # if PET, (chrID, start, end)

        readCur <- read.table( outfileName, sep='\t', nrows=nRow,
            header=FALSE, stringsAsFactors=FALSE, comment.char="",
            colClasses=c("character","numeric","numeric"), check.names=FALSE )
        colnames(readCur) <- c("chrID","start","end")
    } else {
        # if SET, (chrID, start, end, strand)

        readCur <- read.table( outfileName, sep='\t', nrows=nRow,
            header=FALSE, stringsAsFactors=FALSE, comment.char="",
            colClasses=c("character","numeric","numeric","character"),
            check.names=FALSE )
        colnames(readCur) <- c("chrID","start","end","str")
    }

    # match reads for each peak region

    peakRange <- IRanges( start=peakCur[,1], end=peakCur[,2] )
    midpt <- ( readCur[,2] + readCur[,3] ) / 2
    readRange <- IRanges( start=midpt, end=midpt )
    readMatch <- as.matrix( findOverlaps( peakRange, readRange ) )
    matchList <- split( readMatch[,2], readMatch[,1] )

    # check whether there is any peak region without reads

    existInd <- rep( 0, nrow(peakCur) )
    existInd[ unique(readMatch[,1]) ] <- 1

    # update results

    fragSetCur <- vector( "list", length(existInd) )
    cur <- 1
    for ( j in seq_len(length(existInd)) ) {
        if ( existInd[j] == 1 ) {
            fragSetCur[[cur]] <-
                readCur[ matchList[[ as.character(j) ]], -1, drop=FALSE ]
        } else {
            fragSetCur[[cur]] <- matrix( NA )
        }
        cur <- cur + 1
    }

    # update name vector

    nameVecCur <-
        paste( rep(chr,nrow(peakCur)), peakCur[,1], peakCur[,2], sep="_" )

    # check proportion of forward reads for SET data

    if ( PET == FALSE ) {
        nFCur <- length(which( readCur[,4]=="F" ))
        nAllCur <- nrow(readCur)
    } else {
        nFCur <- nAllCur <- NA
    }

    # calculate fragment length for PET data

    if ( PET == TRUE ) {
        fragLenCur <- readCur[,3] - readCur[,2] + 1
    } else {
        fragLenCur <- NA
    }

    return( list( fragSetCur=fragSetCur, nameVecCur=nameVecCur,
        peakChrCur=rep(chr,nrow(peakCur)),
        peakStartCur=peakCur[,1], peakEndCur=peakCur[,2],
        nFCur=nFCur, nAllCur=nAllCur, fragLenCur=fragLenCur ) )
}