Running replicates statistics on bins and bed files

.github/workflows/R-CMD-check.yaml

@@ -67,7 +67,7 @@ jobs:
       - name: Manually install BioConductor dependencies
         run: |
           install.packages("BiocManager")
-          BiocManager::install(c("rtracklayer", "GenomeInfoDb", "GenomicRanges", "BSgenome.Mmusculus.UCSC.mm9", "BSgenome.Hsapiens.UCSC.hg38"))
+          BiocManager::install(c("rtracklayer", "GenomeInfoDb", "GenomicRanges", "BSgenome.Mmusculus.UCSC.mm9", "BSgenome.Hsapiens.UCSC.hg38", "DESeq2"))
         shell: Rscript {0}
         working-directory: elsasserlib
 

elsasserlib/DESCRIPTION

@@ -1,7 +1,7 @@
 Package: elsasserlib
 Type: Package
 Title: General utilities used within Elsasser lab
-Version: 1.0.9
+Version: 1.1.0
 Authors@R: c( 
   person("Carmen", "Navarro", 
          email = "carmen.navarro@scilifelab.se",
@@ -41,7 +41,8 @@ Imports:
   scales,
   RColorBrewer,
   RCurl,
-  pheatmap
+  pheatmap,
+  DESeq2
 Encoding: UTF-8
 LazyData: true
 RoxygenNote: 7.1.1

elsasserlib/NAMESPACE

@@ -2,7 +2,10 @@
 
 export(build_bins)
 export(bw_bed)
+export(bw_bed_diff_analysis)
 export(bw_bins)
+export(bw_bins_diff_analysis)
+export(bw_granges_diff_analysis)
 export(bw_profile)
 export(mean_ratio_norm)
 export(palette_categorical)
@@ -19,6 +22,12 @@ export(trim_quantile)
 import(ggplot2)
 importFrom(BSgenome.Hsapiens.UCSC.hg38,BSgenome.Hsapiens.UCSC.hg38)
 importFrom(BSgenome.Mmusculus.UCSC.mm9,BSgenome.Mmusculus.UCSC.mm9)
+importFrom(DESeq2,DESeqDataSetFromMatrix)
+importFrom(DESeq2,`sizeFactors<-`)
+importFrom(DESeq2,estimateDispersions)
+importFrom(DESeq2,estimateSizeFactors)
+importFrom(DESeq2,nbinomWaldTest)
+importFrom(DESeq2,results)
 importFrom(GenomeInfoDb,seqinfo)
 importFrom(GenomeInfoDb,sortSeqlevels)
 importFrom(GenomicRanges,makeGRangesFromDataFrame)
@@ -49,6 +58,7 @@ importFrom(rmarkdown,render)
 importFrom(rtracklayer,BigWigFile)
 importFrom(rtracklayer,import)
 importFrom(rtracklayer,mcols)
+importFrom(stats,complete.cases)
 importFrom(stats,median)
 importFrom(stats,sd)
 importFrom(stringr,str_sort)

elsasserlib/R/bwstats.R

@@ -0,0 +1,139 @@
+#' Run DESeq2 analysis on genome-wide bins
+#'
+#' Runs a DESeq2 analysis on genome-wide bins of a specified bin size.
+#' The particularity of this analysis is that it skips the estimateSizeFactors
+#' step by default, because this is accounted for in the scaling step of
+#' MINUTE-ChIP samples.
+#'
+#' @param bwfiles_c1 Path or array of paths to the bigWig files for first condition.
+#' @param bwfiles_c2 Path or array of paths to the bigWig files for second condition.
+#' @param genome Genome. Available choices are mm9, hg38.
+#' @param bin_size Bin size.
+#' @inheritParams bw_granges_diff_analysis
+#' @return a DESeqResults object as returned by DESeq2::results function
+#' @export
+bw_bins_diff_analysis <- function(bwfiles_c1,
+                                  bwfiles_c2,
+                                  label_c1,
+                                  label_c2,
+                                  bin_size = 10000,
+                                  genome = "mm9",
+                                  estimate_size_factors = FALSE) {
+
+  bins_c1 <- bw_bins(bwfiles_c1, genome = genome, bin_size = bin_size)
+  bins_c2 <- bw_bins(bwfiles_c2, genome = genome, bin_size = bin_size)
+
+  bw_granges_diff_analysis(bins_c1, bins_c2, label_c1, label_c2,
+                           estimate_size_factors = estimate_size_factors)
+}
+
+#' Run DESeq2 analysis on bed file
+#'
+#' Runs a DESeq2 analysis on a set of loci specified in a BED file.
+#' The particularity of this analysis is that it skips the estimateSizeFactors
+#' step by default, because this is accounted for in the scaling step of
+#' MINUTE-ChIP samples.
+#'
+#' @param bwfiles_c1 Path or array of paths to the bigWig files for first condition.
+#' @param bwfiles_c2 Path or array of paths to the bigWig files for second condition.
+#' @param bedfile BED file for locus specific analysis.
+#' @inheritParams bw_granges_diff_analysis
+#' @return a DESeqResults object as returned by DESeq2::results function
+#' @export
+bw_bed_diff_analysis <- function(bwfiles_c1,
+                                 bwfiles_c2,
+                                 bedfile,
+                                 label_c1,
+                                 label_c2,
+                                 estimate_size_factors = FALSE) {
+
+  loci_c1 <- bw_bed(bwfiles_c1, bedfile = bedfile)
+  loci_c2 <- bw_bed(bwfiles_c2, bedfile = bedfile)
+
+  bw_granges_diff_analysis(loci_c1, loci_c2, label_c1, label_c2,
+                           estimate_size_factors = estimate_size_factors)
+}
+
+
+#' Compute DESeq2 differential analysis on GRanges objects
+#'
+#' Runs a DESeq2 analysis on loci specified on GRanges objects.
+#' The particularity of this analysis is that it skips the estimateSizeFactors
+#' step by default, because this is accounted for in the scaling step of
+#' MINUTE-ChIP samples.
+#'
+#' @param granges_c1 GRanges object containing the values for condition 1.
+#' @param granges_c2 GRanges object containing the values for condition 2.
+#'     Note that these objects must correspond to the same loci.
+#' @param label_c1 Condition name for condition 1.
+#' @param label_c2 Condition name for condition 2.
+#' @param estimate_size_factors If TRUE, normal DESeq2 procedure is done. Set it
+#'     to true to analyze non-MINUTE data.
+#' @importFrom DESeq2 DESeqDataSetFromMatrix estimateDispersions nbinomWaldTest `sizeFactors<-` results estimateSizeFactors
+#' @return a DESeqResults object as returned by DESeq2::results function
+#' @export
+bw_granges_diff_analysis <- function(granges_c1,
+                                     granges_c2,
+                                     label_c1,
+                                     label_c2,
+                                     estimate_size_factors = FALSE) {
+
+  # Bind first, get numbers after (drop complete cases separately could cause error)
+  granges_c1 <- sortSeqlevels(granges_c1)
+  granges_c1 <- sort(granges_c1)
+
+  granges_c2 <- sortSeqlevels(granges_c2)
+  granges_c2 <- sort(granges_c2)
+
+  cts_df <- cbind(data.frame(granges_c1), mcols(granges_c2))
+
+  cts <- get_nreads_columns(cts_df[, 6:ncol(cts_df)])
+
+  condition_labels <- c(rep(label_c1, length(mcols(granges_c1))),
+                        rep(label_c2, length(mcols(granges_c2))))
+
+  coldata <- data.frame(colnames(cts), condition = condition_labels)
+
+  dds <- DESeqDataSetFromMatrix(countData = cts,
+                                colData = coldata,
+                                design = ~ condition)
+
+
+  if (estimate_size_factors == TRUE) {
+    dds <- estimateSizeFactors(dds)
+  }
+  else {
+    # Since files are scaled, we do not want to estimate size factors, so give it
+    # an array of ones
+    sizeFactors(dds) <- c(rep(1, ncol(cts)))
+  }
+
+  dds <- estimateDispersions(dds)
+  dds <- nbinomWaldTest(dds)
+
+  results(dds)
+}
+
+
+#' Get values in a data frame object as round numeric values in a matrix.
+#'
+#' This is an auxiliary function for stats. It drops NAs or NaN values, only
+#' complete cases are used.
+#'
+#' @param df Target data frame
+#' @param length_factor Scaling factor to multiply coverage values by.
+#'
+#' @return An integer matrix
+#' @importFrom stats complete.cases
+get_nreads_columns <- function(df, length_factor = 1000) {
+  # TODO: Consider whether to multiply by locus length. For bins analysis
+  # this should not affect results, but for genes or loci of different length
+  # it might. Since we skip the size factor step, we may bias the results?
+  # So right now it's only fragment length
+  cts <- as.matrix(df)
+  cts <- as.matrix(cts[complete.cases(cts),])
+  cts <- round(cts*length_factor)
+  cts
+}
+
+

elsasserlib/README.Md

@@ -17,7 +17,8 @@ dependencies before running the installation:
         'GenomicRanges',
         'rtracklayer',
         'BSgenome.Mmusculus.UCSC.mm9',
-        'BSgenome.Hsapiens.UCSC.hg38'))
+        'BSgenome.Hsapiens.UCSC.hg38',
+        'DESeq2'))
 
 Then you can install directly from this GitHub repository: