Mercurial > repos > davidvanzessen > shm_csr
view baseline/filter.r @ 66:43a1aa648537 draft
Uploaded
author | davidvanzessen |
---|---|
date | Thu, 07 Dec 2017 03:44:38 -0500 |
parents | c6dd3215ebe0 |
children |
line wrap: on
line source
arg = commandArgs(TRUE) summaryfile = arg[1] gappedfile = arg[2] selection = arg[3] output = arg[4] print(paste("selection = ", selection)) summarydat = read.table(summaryfile, header=T, sep="\t", fill=T, stringsAsFactors=F, quote = "") gappeddat = read.table(gappedfile, header=T, sep="\t", fill=T, stringsAsFactors=F, quote = "") fix_column_names = function(df){ if("V.DOMAIN.Functionality" %in% names(df)){ names(df)[names(df) == "V.DOMAIN.Functionality"] = "Functionality" print("found V.DOMAIN.Functionality, changed") } if("V.DOMAIN.Functionality.comment" %in% names(df)){ names(df)[names(df) == "V.DOMAIN.Functionality.comment"] = "Functionality.comment" print("found V.DOMAIN.Functionality.comment, changed") } return(df) } gappeddat = fix_column_names(gappeddat) #dat = data.frame(merge(gappeddat, summarydat, by="Sequence.ID", all.x=T)) dat = cbind(gappeddat, summarydat$AA.JUNCTION) colnames(dat)[length(dat)] = "AA.JUNCTION" dat$VGene = gsub("^Homsap ", "", dat$V.GENE.and.allele) dat$VGene = gsub("[*].*", "", dat$VGene) dat$DGene = gsub("^Homsap ", "", dat$D.GENE.and.allele) dat$DGene = gsub("[*].*", "", dat$DGene) dat$JGene = gsub("^Homsap ", "", dat$J.GENE.and.allele) dat$JGene = gsub("[*].*", "", dat$JGene) print(str(dat)) dat$past = do.call(paste, c(dat[unlist(strsplit(selection, ","))], sep = ":")) dat = dat[!duplicated(dat$past), ] print(paste("Sequences remaining after duplicate filter:", nrow(dat))) dat = dat[dat$Functionality != "No results" & dat$Functionality != "unproductive",] print(paste("Sequences remaining after functionality filter:", nrow(dat))) print(paste("Sequences remaining:", nrow(dat))) write.table(x=dat, file=output, sep="\t",quote=F,row.names=F,col.names=T)