R/peakCalling.R

In ZW-xjtlu/exomePeak2: Peak Calling and differential analysis for MeRIP-Seq

## Peak calling with MeRIP-Seq.
##
## return the peaks in GRangesList format.
##
peakCalling <- function(bam_IP,
                        bam_input,
                        txdb = NULL,
                        genome = NULL,
                        bin_size = 25,
                        step_size = 25,
                        fragment_length = 100,
                        strandness = c("unstrand", "1st_strand", "2nd_strand"),
                        gff = NULL,
                        test_method = c("Poisson", "DESeq2"),
                        p_cutoff = 1e-10,
                        plot_gc = FALSE,
                        parallel = 1,
                        motif_based = FALSE,
                        motif_sequence = "DRACH",
                        fig_dir = "exomePeak2_output",
                        mode = c("exon","full_transcript","whole_genome"),
                        confounding_factor = NULL){
  #require(GenomicRanges)
  #require(SummarizedExperiment)
  #require(DESeq2)

  #Check input validity
  strandness <- match.arg(strandness)
  test_method <- match.arg(test_method)
  stopifnot(fragment_length > 0 & bin_size > 0 & step_size > 0)

  if(is.null(gff) & is.null(txdb)){
    stop("Please at least provide one among gff and TxDb for transcript annotation.")
  }

  if(is.null(txdb) & !is.null(gff)){
    txdb <- makeTxDbFromGFF(gff)
  }

  if(!(all(file.exists(bam_IP)) & all(file.exists(bam_input)))){
    stop("At least one bam file directories provided cannot be found, please check it.")
  }

  if(is.null(genome) & motif_based){
    stop("Motif cannot be extracted without providing the 'genome' argument, please try to find the reference genome.")
  }

  if(is.null(genome)){
    warning("Reference genome not provided, GC content bias is left uncorrected.")
  }

  #Extract bins for count

  message("Extract bin features ... ", appendLF = FALSE)
  exByGene  <- exonsByiGenes(txdb) %>% quiet
  if(!motif_based){
    peakBins <- exonicBins(exByGene, bin_size, step_size) %>% quiet
  }else{
    peakBins <- sampleSequence(motif_sequence, exByGene, genome) %>% quiet
    peakBins <-  resize(peakBins, 1, fix = "center") %>% quiet
  }
  mcols(peakBins) <- NULL
  if(is(peakBins, "GRanges")) peakBins <- split(peakBins, seq_along(peakBins)) %>% quiet
  message("OK")

  #Count the bam files
  bam_dirs <- c(bam_IP, bam_input)

  message("Count reads on bin features ... ", appendLF = FALSE)
  se <- featuresCounts(peakBins, bam_dirs, strandness, parallel) %>% quiet
  rm(bam_dirs)
  message("OK")

  #Annotate SummarizedExperiment
  se$IP_input <- rep(c("IP", "input"), c(length(bam_IP), length(bam_input)))
  rm(peakBins)

  #Identify Backgrounds
  message("Identify background features ... ", appendLF = FALSE)
  se <- classifyBackground(se) %>% quiet
  message("OK")

  #Estimate sample size factors
  message("Estimate sample sepecific size factors from the background ... ", appendLF = FALSE)
  se <- estimateColumnFactors(se) %>% quiet
  message("OK")

  if(!is.null(genome)){
  #Calculate GC contents
  message("Calculate bin GC contents on exons ... ", appendLF = FALSE)
  se <- calculateGCcontents(se, fragment_length, exByGene, genome) %>% quiet
  message("OK")

  #Fit GC content biases
  message("Fit GC curves with smoothing splines ... ", appendLF = FALSE)
  se <- fitBiasCurves(se) %>% quiet
  message("OK")

  #Estimate matrix correction factors
  message("Calculate offset matrix ... ", appendLF = FALSE)
  se <- estimateMatrixFactors(se) %>% quiet
  message("OK")

  ## Plot GC bias fits
  if(plot_gc) plotGCbias(se, fig_dir) %>% quiet

  }else{
  #Assign matrix correction factors without GC offsets
  se <- estimateMatrixFactors(se) %>% quiet
  }

  #Filter low count rows if not exon mode
  if(mode %in% c("full_transcript","whole_genome")){
    se <- se[rowMeans(assay(se)) >= 5,]
  }
  
  #DESeq2 peak calling
  message("Detect peaks with GLM ... ", appendLF = FALSE)
  peaks <- callPeaks(se, txdb, test_method, p_cutoff, exByGene, bin_size, motif_based, confounding_factor) %>% quiet
  message("OK")

  return(peaks)
}