#!/usr/bin/env Rscript # Copyright (c) 2013 Daniel S. Standage, released under MIT license # # expr-dist: plot distributions of expression values before and after # normalization; visually confirm that normalization worked # as expected # # Program input is a matrix of expression values, each row corresponding to a # molecule (gene, transcript, etc) and each row corresponding to that molecule's # expression level or abundance. The program expects the rows and columns to be # named, and was tested primarily on output produced by the # 'rsem-generate-data-matrix' script distributed with the RSEM package. # # The program plots the distributions of the logged expression values by sample # as provided, then normalizes the values, and finally plots the distribution of # the logged normalized expression values by sample. The expectation is that all # samples' distributions will have a similar shape but different medians prior # to normalization, and that post normalization they will all have an identical # median to facilitate cross-sample comparison. # MedianNorm function borrowed from the EBSeq library version 1.1.6 # See http://www.bioconductor.org/packages/devel/bioc/html/EBSeq.html MedianNorm <- function(data) { geomeans <- exp( rowMeans(log(data)) ) apply(data, 2, function(cnts) median((cnts/geomeans)[geomeans > 0])) } library("getopt") print_usage <- function(file=stderr()) { cat(" expr-dist: see source code for full description Usage: expr-dist [options] < expr-matrix.txt Options: -h|--help: print this help message and exit -o|--out: STRING prefix for output files; default is 'expr-dist' -r|--res: INT resolution (dpi) of generated graphics; default is 150 -t|--height: INT height (pixels) of generated graphics; default is 1200 -w|--width: INT width (pixels) of generated graphics; default is 1200 -y|--ylim: REAL the visible range of the Y axis depends on the first distribution plotted; if other distributions are getting cut off, use this setting to override the default\n\n") } spec <- matrix( c("help", 'h', 0, "logical", "out", 'o', 1, "character", "res", 'r', 1, "integer", "height", 't', 1, "integer", "width", 'w', 1, "integer", "ylim", 'y', 1, "double"), byrow=TRUE, ncol=4) opt <- getopt(spec) if(!is.null(opt$help)) { print_usage(file=stdout()) q(status=1) } if(is.null(opt$height)) { opt$height <- 1200 } if(is.null(opt$out)) { opt$out <- "expr-dist" } if(is.null(opt$res)) { opt$res <- 150 } if(is.null(opt$width)) { opt$width <- 1200 } if(!is.null(opt$ylim)) { opt$ylim <- c(0, opt$ylim) } # Load data, determine number of samples data <- read.table(file("stdin"), header=TRUE, sep="\t", quote="") nsamp <- dim(data)[2] - 1 data <- data[,1:nsamp+1] # Plot distribution of expression values before normalization outfile <- sprintf("%s-median.png", opt$out) png(outfile, height=opt$height, width=opt$width, res=opt$res) h <- hist(log(data[,1]), plot=FALSE) plot(h$mids, h$density, type="l", col=rainbow(nsamp)[1], main="", xlab="Log expression value", ylab="Proportion of molecules", ylim=opt$ylim) for(i in 2:nsamp) { h <- hist(log(data[,i]), plot=FALSE) lines(h$mids, h$density, col=rainbow(nsamp)[i]) } devnum <- dev.off() # Normalize by median size.factors <- MedianNorm(data.matrix(data)) data.norm <- t(apply(data, 1, function(x){ x / size.factors })) # Plot distribution of normalized expression values outfile <- sprintf("%s-median-norm.png", opt$out) png(outfile, height=opt$height, width=opt$width, res=opt$res) h <- hist(log(data.norm[,1]), plot=FALSE) plot(h$mids, h$density, type="l", col=rainbow(nsamp)[1], main="", xlab="Log normalized expression value", ylab="Proportion of molecules", ylim=opt$ylim) for(i in 2:nsamp) { h <- hist(log(data.norm[,i]), plot=FALSE) lines(h$mids, h$density, col=rainbow(nsamp)[i]) } devnum <- dev.off()