diff plot_comparative_clustering_summary.R @ 17:d14b68e9fd1d draft

Uploaded - new tools added
author petr-novak
date Wed, 28 Apr 2021 08:37:20 +0000
parents
children 5a05925340b0
line wrap: on
line diff
--- /dev/null	Thu Jan 01 00:00:00 1970 +0000
+++ b/plot_comparative_clustering_summary.R	Wed Apr 28 08:37:20 2021 +0000
@@ -0,0 +1,304 @@
+#!/usr/bin/env Rscript
+library(optparse)
+## TODO - add scale to legend!
+twenty_colors = c(
+  '#e6194b', '#3cb44b', '#ffe119', '#4363d8', '#f58231',
+  '#911eb4', '#46f0f0', '#f032e6', '#bcf60c', '#fabebe',
+  '#008080', '#e6beff', '#9a6324', '#fffac8', '#800000',
+  '#aaffc3', '#808000', '#ffd8b1', '#000075', "#000000"
+)
+
+get_color = function(classification, size){
+  ## 20 of unique colors, first is black
+  unique_colors = twenty_colors[1:opt$number_of_colors]
+  Ncol = length(unique_colors)
+  ## rest wil be grey:
+  grey_color = "#a9a9a9"
+  ## unique repeats without All
+  include = !classification %in% "All"
+  unique_repeats = names(c(sort(by(size[include], INDICES = classification[include], FUN = sum), decreasing = TRUE)))
+  color_table = unique_colors[1:min(Ncol,length(unique_repeats))]
+  names(color_table) = unique_repeats[1:min(Ncol,length(unique_repeats))]
+  color = rep(grey_color, length(classification))
+  names(color) = classification
+  for (ac in names(color_table)){
+    color[names(color) %in% ac] = color_table[ac]
+  }
+  return(color)
+}
+
+
+make_legend = function(color){
+  ## simplify description:
+  names(color) = gsub(".+/","",names(color))
+  description = sapply(split(names(color), color), function(x) paste(unique(x), collapse=";"))
+  description = gsub(".+;.+", "Other", description)
+  description = gsub("All", "Other", description)
+  if ("Other" %in% description & length(description) > 1){
+    description = c(description[! description %in% "Other"], description[description %in% "Other"])
+  }
+  ord = order(factor(names(description), levels = twenty_colors))
+  legend_info = list(name = gsub("All", "NA", description)[ord], color = names(description)[ord])
+}
+
+plot_rect_map = function(read_counts,cluster_annotation, output_file,GS, RL, Xcoef=1,Ycoef=1){
+  ## read_counts : correspond to COMPARATIVE_ANALYSIS_COUNTS.csv
+  ## cluster annotation : CLUSTER_TABLE.csv
+  counts = read.table(read_counts,header=TRUE,as.is=TRUE)
+  input_read_counts = unlist(read.table(read_counts, nrows = 1, comment.char = "",sep="\t")[-(1:2)])
+
+  counts_file_valid = ncol(counts) == (length(input_read_counts) + 2) & all(colnames(input_read_counts)[1:2]==c("cluster", "supercluster"))
+  ## find which line is header
+  header_line = grep(".*Cluster.*Supercluster.*Size", readLines(cluster_annotation))
+  annot = read.table(cluster_annotation, sep="\t",header=TRUE,as.is=TRUE, skip = header_line - 1)
+  ## validate
+  annot_file_valid = all(colnames(annot)==c("Cluster","Supercluster","Size","Size_adjusted","Automatic_annotation","TAREAN_annotation","Final_annotation"))
+
+
+  if (!annot_file_valid | !counts_file_valid){
+    pdf(output_file)
+    plot.new()
+    text(0.5,0.5,"Input is not valid, check input files!")
+    dev.off()
+    stop("Input files are not valid!")
+  }
+  print(annot_file_valid)
+  print(counts_file_valid)
+  ## remove counts which are not in annotation - only clusters in annot will be plotted!
+  counts = counts[annot$Cluster,]
+  N = nrow(annot)
+
+  counts_automatic = counts
+  annot_automatic = annot
+  input_read_counts_automatic = input_read_counts
+  # remove organelar and contamination if required make count correction
+  if (opt$nuclear_only){
+    exclude=grep("contamination|organelle",annot$Automatic_annotation)
+    if (length(exclude)>0){
+      counts_automatic = counts[-exclude, , drop=FALSE]
+      annot_automatic = annot[-exclude, ,drop=FALSE]
+      input_read_counts_automatic = input_read_counts - colSums(counts[exclude,-c(1:2) , drop=FALSE])
+    }
+  }
+  color_auto = get_color(annot_automatic$Automatic_annotation, annot_automatic$Size)
+
+  legend_info = make_legend(color_auto)
+  params = list(Automatic_annotation = list(
+                  color =  color_auto,
+                  legend = legend_info,
+                  counts = counts_automatic,
+                  annot = annot_automatic,
+                  input_read_counts = input_read_counts_automatic
+                )
+                )
+
+
+  if (!is.null(annot$Final_annotation)){
+
+    ## column with manual annotation exist - check if correct
+    if (any(annot$Final_annotation %in% "" | any(is.na(annot$Final_annotation)))){
+      message("Final annotation is not complete, skipping")
+    }else{
+
+      counts_manual = counts
+      annot_manual = annot
+      input_read_counts_manual = input_read_counts
+      ## correction must be done idependetly in case manual and automatic classification differ in annotation
+      if (opt$nuclear_only){
+        exclude=grep("contamination|organelle",annot$Final_annotation)
+        if (length(exclude)>0){
+          counts_manual = counts[-exclude, , drop=FALSE]
+          annot_manual = annot[-exclude, ,drop=FALSE]
+          input_read_counts_manual = input_read_counts - colSums(counts[exclude,-c(1:2) , drop=FALSE])
+        }
+      }
+      ## append
+      color_manual = get_color(annot_manual$Final_annotation, annot_manual$Size)
+      legend_info_manual = make_legend(color_manual)
+
+      params$Final_annotation = list(
+        color =  color_manual,
+        legend = legend_info_manual,
+        counts = counts_manual,
+        annot = annot_manual,
+        input_read_counts = input_read_counts_manual
+
+      )
+    }
+  }
+
+  ## set size of pdf output
+  wdth = (3 + N*0.03 ) * Xcoef
+  hgt = (2.2 + ncol(counts)*0.20) * Ycoef
+  if (!any(is.na(GS))){
+    hgt = hgt + ncol(counts)*0.20 * Ycoef
+  }
+  ptsize = round((wdth*hgt)^(1/4))*5
+  
+
+  pdf(output_file, width=wdth,height=hgt, pointsize = ptsize)  # was 50
+  ## originaly - printing of both automatic and final annotation
+  ## now - print only final_annotation if available
+  if  (length(params) == 2){
+    ## remove automatic
+    params$Automatic_annotation = NULL
+  }
+  ## 
+
+  for (j in seq_along(params)){
+    Nclust = nrow(params[[j]]$annot)
+    ##prepare matrix for plotting
+    M = as.matrix(params[[j]]$counts[1:Nclust,-(1:2)])
+    rownames(M) = paste0("CL",rownames(M))
+    Mn1=(M)/apply(M,1,max)
+    Mn2=M/max(M)
+    ord1 = hclust(dist(Mn1),method = "ward.D")$order
+    ord2 = hclust(dist(t(Mn2)))$order
+
+    ploting_area_width = 3 + log10(Nclust)*3
+    ploting_area_sides = 1.5
+    legend_width = 3
+    title_height = 0.5
+    if (any(is.na(GS))){
+      layout(matrix(c(0,0,0,3,0,2,0,3,0,1,0,3),ncol=4,byrow = TRUE),
+             width=c(ploting_area_sides,ploting_area_width,ploting_area_sides, legend_width),
+             height=c(title_height, 3,ncol(M)*0.8))
+    }else{
+      ## extra row for legends
+
+
+      layout(matrix(c(0,0,0,3,0,2,0,3,0,1,0,3,0,0,0,4),ncol=4,byrow = TRUE),
+             width=c(ploting_area_sides,ploting_area_width,ploting_area_sides, legend_width),
+             height=c(title_height, 3,ncol(M)*0.8,ncol(M)*0.8 ))
+    }
+
+
+    par(xaxs='i', yaxs = 'i')
+    par(las=2,mar=c(4,0,0,0),cex.axis=0.5)
+
+    if (any(is.na(GS))){
+      rectMap(Mn2[ord1,ord2],scale.by='row',col=params[[j]]$color[ord1], grid=TRUE)
+    }else{
+      # use genomic sizes
+      Mn3 = t(t(M) * (GS[colnames(M),] / params[[j]]$input_read_counts))[ord1,ord2]
+      ## rescale
+      MaxGS = max(Mn3)
+      Mn3 = Mn3/max(Mn3)
+      rectMap(Mn3,scale.by='none',col=params[[j]]$color[ord1], grid=TRUE)
+    }
+    par(las=2,mar=c(1,0,1,0), mgp = c(2,1,0))
+    barplot(params[[j]]$annot$Size[ord1], col = 1)
+    mtext(side = 2, "Cluster size", las = 3, line = 2.5, cex = 0.5)
+    mtext(side=3, names(params)[j], las=0, line=1)
+    plot.new()
+    legend("topleft", col= params[[j]]$legend$color, legend=params[[j]]$legend$name, pch=15, cex=0.7, bty="n", pt.cex=1)
+  }
+
+  if (!any(is.na(GS))){
+    ## plot GS scale
+    par(xaxs='i', yaxs = 'i')
+    print(log(nrow(Mn3)))
+    par(las=2,mar=c(4,0,0,log(nrow(Mn3))),cex.axis=0.5) # same par as recplot above to keep the scale
+    Mn3scale = Mn3
+    Mn3scale[,] = 0
+    colnames(Mn3scale)=rep("", ncol(Mn3scale))
+    rownames(Mn3scale)=rep("", nrow(Mn3scale))
+    Mn3scale[,1] = seq(0,1, length.out = nrow(Mn3))
+    rectMap(Mn3scale,scale.by='none',col="grey", grid=FALSE, boxlab="", draw_box=FALSE, center=FALSE)
+    slabels = pretty(c(0,MaxGS), n = 10)
+    sat = slabels/MaxGS * nrow(Mn3scale)
+    axis(side=1, at= sat, labels = slabels, line = 0)
+    mtext(side = 1, text = "Repeat abundance", las=1, line=2.5,cex=0.4)
+    mtext(side = 2, text = "Rectangle\n height", las=1, line=2,cex=0.4, at=1)
+
+    axis(2, at=c(0.5, 1, 1.5), labels=c(0,0.5,1),line=0)
+  }
+  st = dev.off()
+}
+
+rectMap=function(x,scale.by='row',col=1,xlab="",ylab="",grid=TRUE,axis_pos=c(1,4),boxlab = "Cluster Id", cexx=NULL,cexy=NULL, draw_box=TRUE, center=TRUE){
+  if (scale.by=='row'){
+                                        #x=(x)/rowSums(x)
+    x=(x)/apply(x,1,sum)
+  }
+  if (scale.by=='column'){
+    x=t(t(x)/apply(x,2,max))
+  }
+  nc=ncol(x)
+  nr=nrow(x)
+  coords=expand.grid(1:nr,1:nc)
+  plot(coords[,1],coords[,2],type='n',axes=F,xlim=range(coords[,1])+c(-.5,.5),ylim=range(coords[,2])+c(-.5,.5),xlab=xlab,ylab=ylab)
+  axis(axis_pos[1],at=1:nr,labels=rownames(x),lty=0,tick=FALSE,line=0,cex.axis=0.5/log10(nr))
+  axis(axis_pos[2],at=1:nc,labels=colnames(x),lty=0,tick=FALSE,las=2,line=0 ,hadj=0, cex.axis=0.7)
+  axis(2,at=1:nc,labels=colnames(x),lty=0,tick=FALSE,las=2,line=0 ,hadj=1, cex.axis=0.7)
+
+  mtext(side = 1, boxlab, las=1, line = 3, cex = 0.5)
+  line = 1.5 + log10(nr)
+  #mtext(side = 2, "Proportions of individual samples", las =0, line = line, cex = 0.5)
+  s=x/2
+  w = c(x)/2
+  if(center){
+    rect(coords[,1]-0.5,coords[,2]-s,coords[,1]+0.5,coords[,2]+s,col=col,border=NA)
+  }else{
+    rect(coords[,1]-0.5,coords[,2]-0.5,coords[,1]+0.5,coords[,2]+x-0.5,col=col,border=NA)
+  }
+  if (grid){
+    abline(v=0:(nr)+.5,h=0:(nc)+.5,lty=2,col="#60606030",lwd=0.2)
+  }
+  if(draw_box){
+    box(col="#60606030",lty=2, lwd=0.2)
+  }
+}
+
+ option_list <- list( 
+  make_option(c("-c", "--cluster_table"), default=NA, type = "character",
+              help="file from RepeatExplorer2 clustering - CLUSTER_TABLE.csv"),
+
+  make_option(c("-m", "--comparative_counts"),default = NA,type = "character",
+              help="file from RepeatExplorer2 output - COMPARATIVE_ANALYSIS_COUNTS.csv"),
+
+  make_option(c("-o", "--output"), type="character",
+              default="comparative_analysis_summary.pdf",
+              help="File name for output figures (pdf document)"),
+  make_option(c("-N", "--number_of_colors"), type="integer", default=10,
+              help="Number of unique colors used from plotting (2-20, default is 10)"),
+
+  make_option(c("-g", "--genome_size"),default = NA,type = "character",
+              help="file from genome sizes of species provided in tab delimited file in the format:
+
+                   species_code1   GenomeSize1
+                   species_code2   GenomeSize2
+                   species_code3   GenomeSize3
+                   species_code4   GenomeSize4
+
+                provide the same codes for species as in file COMPARATIVE_ANALYSIS_COUNTS.csv. The use of genome
+                sizes file imply the --nuclear_only option. If genome sizes are used, genomic abundance scale is added.
+    "),
+  make_option(c("-n", "--nuclear_only"), default = FALSE, type="logical",
+              action = "store_true",
+              help="remove all non-nuclear sequences (organelle and contamination). ")
+)
+
+
+opt = parse_args(OptionParser(option_list = option_list))
+
+if (any(is.na(c(opt$cluster_table, opt$comparative_counts)))){
+  message("\nBoth files: CLUSTER_TABLE.csv and COMPARATIVE_ANALYSIS_COUNTS.csv must be provided\n")
+  q()
+}
+
+if (!opt$number_of_colors %in% 1:20){
+  message("number of color must be in range 1..20")
+  stop()
+}
+
+if (!is.na(opt$genome_size)){
+  GS = read.table(opt$genome_size, header=FALSE, as.is=TRUE, row.names = 1)
+  opt$nuclear_only=TRUE
+}else{
+  GS = NA
+  RL = NA
+}
+
+plot_rect_map(opt$comparative_counts, opt$cluster_table, opt$output, GS, RL)
+