# This script adjusts an ART error profile to increase the cumulative error # rate to a given value. ## Optimization of multiplier of counts vector to generate ## a distribution with (approximately) a given error rate q2p <- function(q) 10^(-q/10) cum_err <- function(p, counts) { s <- sum(counts) sum(p * counts) / s } eq <- function(a, b, tol) { return (abs(a-b) <= tol) } new_err <- function(d, p, counts, target.err) { total <- sum(counts) new.counts <- counts + (total * d) abs(cum_err(p, (new.counts / sum(new.counts)) * total) - target.err) } find_new_counts <- function(p, counts, target.err, rescale=TRUE, d0=target.err*0.1) { res <- optim(par=d0, fn=new_err, p=p, counts=counts, target.err=target.err, method = "BFGS") new.counts <- counts + (sum(counts) * res$par) if (rescale) { m <- sum(counts) / sum(new.counts) new.counts <- new.counts * m } list(counts=new.counts, res=res) } profile.error <- function(path) { if (endsWith(path, ".gz")) { lines <- readLines(gzfile(path)) } else { lines <- readLines(path) } comments <- grep("^#", lines, perl=T) if (length(comments) > 0) { lines <- lines[-comments] } num <- NULL denom <- NULL for (i in seq(1, length(lines), 2)) { q <- unlist(strsplit(lines[i], "\t")) if (length(q) < 3) { next } h <- q[1:2] q <- as.integer(q[3:length(q)]) p <- sapply(q, q2p) counts <- unlist(strsplit(lines[i+1], "\t")) stopifnot(counts[1] == h[1] && counts[2] == h[2]) counts <- as.numeric(counts[3:length(counts)]) stopifnot(length(counts) == length(p)) num <- c(num, p*counts) denom <- c(denom, counts) } as.numeric(sum(num)) / as.numeric(sum(denom)) } ## Generate a new error profile adjust.profile <- function(orig.path, new.path, target.err, rescale=TRUE) { if (endsWith(orig.path, ".gz")) { lines <- readLines(gzfile(orig.path)) } else { lines <- readLines(orig.path) } comments <- grep("^#", lines, perl=T) if (length(comments) > 0) { lines <- lines[-comments] } n.lines <- length(lines) read.length <- n.lines / 12 new.lines <- lines process.lines <- function(i) { q <- unlist(strsplit(lines[i], "\t")) h <- q[1:2] q <- as.integer(q[3:length(q)]) p <- sapply(q, q2p) counts <- unlist(strsplit(lines[i+1], "\t")) stopifnot(counts[1] == h[1] && counts[2] == h[2]) counts <- as.numeric(counts[3:length(counts)]) stopifnot(length(counts) == length(p)) new.counts <- find_new_counts(p, counts, target.err, rescale) list(h=h, q=q, p=p, counts=new.counts$counts, line=paste0(c(h, round(new.counts$counts)), collapse="\t")) } new.N <- rep(NA, read.length) for (ii in 1:read.length) { i <- (ii*2)-1 result <- process.lines(i) new.lines[i+1] <- result$line w <- which(result$q == 2) if (length(w) > 0) { new.N[ii] <- result$counts[w[1]] } } end <- read.length * 2 * 5 for (i in seq((read.length * 2) + 1, end, 2)) { result <- process.lines(i) new.lines[i+1] <- result$line } for (i in 1:read.length) { if (!is.na(new.N[i])) { new.lines[end+(i*2)] <- paste0(c('N', i-1, new.N[i]), collapse="\t") } } if (endsWith(new.path, ".gz")) { out <- gzfile(new.path, "w") } else { out <- file(new.path, "w") } writeLines(new.lines, out) close(out) } parser <- ArgumentParser() parser$add_argument( "-g", "--generate", action="store_true", default=FALSE, help="Generate new profile.") parser$add_argument( "-c", "--calculate", action="store_true", default=FALSE, help="Calculate error rate on existing profile.") parser$add_argument( "-p", "--profile", help="The source ART error profile.") parser$add_argument( "-e", "--error-rate", type="double", action="store_true", default=FALSE, help="Calculate error rate of existing profile.") parser$add_argument( "-o", "--output", default=NULL, help="New profile file.") parser$add_argument( "-t", "--table", default=NULL, help="File where data will be written in tab-delimited format.") args <- parser$parse_args() if (args$generate) { # for some reason, the acutal error rate in simulated reads is about twice # what is expected based on the empirical distribution of quality scores error.rate <- as.numeric(args$error_rate) / 2 adjust.profile(args$profile, args$output, error.rate) actual.error <- profile.error(args$output) print(paste("The acutal error rate of the new profile is", actual.error)) } else if (args$error) { print(paste("The acutal error rate of the profile is", round(profile.error(args$profile), 4))) } if (args$table) { if (!file.exists(args$table)) { write( paste("Profile", "RequestedErrorRate", "ActualErrorRate", sep="\t"), file=args$table) } write( paste(basename(args$profile), error.rate, actual.error, sep="\t"), file=args$table, append=TRUE) }