Mercurial > repos > davidvanzessen > argalaxy_tools
view igblast/igblast.r @ 48:efe0a2b5d163 draft
Uploaded
author | davidvanzessen |
---|---|
date | Tue, 18 Jun 2019 09:23:41 -0400 |
parents | afe85eb6572e |
children | 124b7fd92a3e |
line wrap: on
line source
args <- commandArgs(trailingOnly = TRUE) infile=args[1] outfile=args[2] blasted = read.table(infile, header=T, sep="\t", fill=T, stringsAsFactors=F, comment.char="") blasted$ID = 1:nrow(blasted) blasted$VDJ.Frame = "Out-of-frame" search = blasted$inFrame == "true" & blasted$noStop == "false" if(sum(search) > 0){ blasted[search ,]$VDJ.Frame = "In-frame with stop codon" } search = blasted$inFrame == "true" & blasted$noStop == "true" if(sum(search) > 0){ blasted[search ,]$VDJ.Frame = "In-frame" } blasted$Top.V.Gene = blasted$vSegment blasted$Top.D.Gene = blasted$dSegment blasted$Top.J.Gene = blasted$jSegment blasted$CDR1.Seq = blasted$cdr1aa blasted$CDR1.Length = nchar(blasted$CDR1.Seq) blasted$CDR2.Seq = blasted$cdr2aa blasted$CDR2.Length = nchar(blasted$CDR2.Seq) blasted$CDR3.Seq = blasted$cdr3aa blasted$CDR3.Length = nchar(blasted$CDR3.Seq) blasted$CDR3.Seq.DNA = blasted$cdr3nt blasted$CDR3.Length.DNA = nchar(blasted$CDR3.Seq.DNA) blasted$Strand = "+/-" blasted$CDR3.Found.How = "found" search = blasted$cdr3nt == "" if(sum(search) > 0){ blasted[search,]$CDR3.Found.How = "NOT_FOUND" } blasted$AA.JUNCTION = blasted$CDR3.Seq n = c("X.reads_count", "ID", "VDJ.Frame", "Top.V.Gene", "Top.D.Gene", "Top.J.Gene", "CDR1.Seq", "CDR1.Length", "CDR2.Seq", "CDR2.Length", "CDR3.Seq", "CDR3.Length", "CDR3.Seq.DNA", "CDR3.Length.DNA", "Strand", "CDR3.Found.How", "Functionality", "AA.JUNCTION") n[!(n %in% names(blasted))] blasted = blasted[,c("X.reads_count", "ID", "VDJ.Frame", "Top.V.Gene", "Top.D.Gene", "Top.J.Gene", "CDR1.Seq", "CDR1.Length", "CDR2.Seq", "CDR2.Length", "CDR3.Seq", "CDR3.Length", "CDR3.Seq.DNA", "CDR3.Length.DNA", "Strand", "CDR3.Found.How", "AA.JUNCTION")] names(blasted) = c("frequency.count", "ID", "VDJ Frame", "Top V Gene", "Top D Gene", "Top J Gene", "CDR1 Seq", "CDR1 Length", "CDR2 Seq", "CDR2 Length", "CDR3 Seq", "CDR3 Length", "CDR3 Seq DNA", "CDR3 Length DNA", "Strand", "CDR3 Found How", "AA JUNCTION") #duplicate rows based on frequency.count blasted = blasted[rep(seq_len(nrow(blasted)), blasted$frequency.count),] blasted$ID = 1:nrow(blasted) blasted = blasted[,c("ID", "VDJ Frame", "Top V Gene", "Top D Gene", "Top J Gene", "CDR1 Seq", "CDR1 Length", "CDR2 Seq", "CDR2 Length", "CDR3 Seq", "CDR3 Length", "CDR3 Seq DNA", "CDR3 Length DNA", "Strand", "CDR3 Found How", "AA JUNCTION")] write.table(blasted, outfile, quote=F, sep="\t", row.names=F, col.names=T)