| 1 |
# This script adjusts an ART error profile to increase the cumulative error |
| 2 |
# rate to a given value. |
| 3 |
|
| 4 |
## Optimization of multiplier of counts vector to generate |
| 5 |
## a distribution with (approximately) a given error rate |
| 6 |
q2p <- function(q) 10^(-q/10) |
| 7 |
cum_err <- function(p, counts) { |
| 8 |
s <- sum(counts) |
| 9 |
sum(p * counts) / s |
| 10 |
} |
| 11 |
eq <- function(a, b, tol) { |
| 12 |
return (abs(a-b) <= tol) |
| 13 |
} |
| 14 |
new_err <- function(d, p, counts, target.err) { |
| 15 |
total <- sum(counts) |
| 16 |
new.counts <- counts + (total * d) |
| 17 |
abs(cum_err(p, (new.counts / sum(new.counts)) * total) - target.err) |
| 18 |
} |
| 19 |
find_new_counts <- function(p, counts, target.err, rescale=TRUE, d0=target.err*0.1) { |
| 20 |
res <- optim(par=d0, fn=new_err, p=p, counts=counts, target.err=target.err, method = "BFGS") |
| 21 |
new.counts <- counts + (sum(counts) * res$par) |
| 22 |
if (rescale) { |
| 23 |
m <- sum(counts) / sum(new.counts) |
| 24 |
new.counts <- new.counts * m |
| 25 |
} |
| 26 |
list(counts=new.counts, res=res) |
| 27 |
} |
| 28 |
|
| 29 |
profile.error <- function(path) { |
| 30 |
if (endsWith(path, ".gz")) { |
| 31 |
lines <- readLines(gzfile(path)) |
| 32 |
} |
| 33 |
else { |
| 34 |
lines <- readLines(path) |
| 35 |
} |
| 36 |
comments <- grep("^#", lines, perl=T) |
| 37 |
if (length(comments) > 0) { |
| 38 |
lines <- lines[-comments] |
| 39 |
} |
| 40 |
num <- NULL |
| 41 |
denom <- NULL |
| 42 |
for (i in seq(1, length(lines), 2)) { |
| 43 |
q <- unlist(strsplit(lines[i], "\t")) |
| 44 |
if (length(q) < 3) { |
| 45 |
next |
| 46 |
} |
| 47 |
h <- q[1:2] |
| 48 |
q <- as.integer(q[3:length(q)]) |
| 49 |
p <- sapply(q, q2p) |
| 50 |
counts <- unlist(strsplit(lines[i+1], "\t")) |
| 51 |
stopifnot(counts[1] == h[1] && counts[2] == h[2]) |
| 52 |
counts <- as.numeric(counts[3:length(counts)]) |
| 53 |
stopifnot(length(counts) == length(p)) |
| 54 |
num <- c(num, p*counts) |
| 55 |
denom <- c(denom, counts) |
| 56 |
} |
| 57 |
as.numeric(sum(num)) / as.numeric(sum(denom)) |
| 58 |
} |
| 59 |
|
| 60 |
## Generate a new error profile |
| 61 |
adjust.profile <- function(orig.path, new.path, target.err, rescale=TRUE) { |
| 62 |
if (endsWith(orig.path, ".gz")) { |
| 63 |
lines <- readLines(gzfile(orig.path)) |
| 64 |
} |
| 65 |
else { |
| 66 |
lines <- readLines(orig.path) |
| 67 |
} |
| 68 |
comments <- grep("^#", lines, perl=T) |
| 69 |
if (length(comments) > 0) { |
| 70 |
lines <- lines[-comments] |
| 71 |
} |
| 72 |
n.lines <- length(lines) |
| 73 |
read.length <- n.lines / 12 |
| 74 |
new.lines <- lines |
| 75 |
process.lines <- function(i) { |
| 76 |
q <- unlist(strsplit(lines[i], "\t")) |
| 77 |
h <- q[1:2] |
| 78 |
q <- as.integer(q[3:length(q)]) |
| 79 |
p <- sapply(q, q2p) |
| 80 |
counts <- unlist(strsplit(lines[i+1], "\t")) |
| 81 |
stopifnot(counts[1] == h[1] && counts[2] == h[2]) |
| 82 |
counts <- as.numeric(counts[3:length(counts)]) |
| 83 |
stopifnot(length(counts) == length(p)) |
| 84 |
new.counts <- find_new_counts(p, counts, target.err, rescale) |
| 85 |
list(h=h, q=q, p=p, counts=new.counts$counts, line=paste0(c(h, round(new.counts$counts)), collapse="\t")) |
| 86 |
} |
| 87 |
new.N <- rep(NA, read.length) |
| 88 |
for (ii in 1:read.length) { |
| 89 |
i <- (ii*2)-1 |
| 90 |
result <- process.lines(i) |
| 91 |
new.lines[i+1] <- result$line |
| 92 |
w <- which(result$q == 2) |
| 93 |
if (length(w) > 0) { |
| 94 |
new.N[ii] <- result$counts[w[1]] |
| 95 |
} |
| 96 |
} |
| 97 |
end <- read.length * 2 * 5 |
| 98 |
for (i in seq((read.length * 2) + 1, end, 2)) { |
| 99 |
result <- process.lines(i) |
| 100 |
new.lines[i+1] <- result$line |
| 101 |
} |
| 102 |
for (i in 1:read.length) { |
| 103 |
if (!is.na(new.N[i])) { |
| 104 |
new.lines[end+(i*2)] <- paste0(c('N', i-1, new.N[i]), collapse="\t") |
| 105 |
} |
| 106 |
} |
| 107 |
if (endsWith(new.path, ".gz")) { |
| 108 |
out <- gzfile(new.path, "w") |
| 109 |
} |
| 110 |
else { |
| 111 |
out <- file(new.path, "w") |
| 112 |
} |
| 113 |
writeLines(new.lines, out) |
| 114 |
close(out) |
| 115 |
} |
| 116 |
|
| 117 |
parser <- ArgumentParser() |
| 118 |
parser$add_argument( |
| 119 |
"-g", "--generate", action="store_true", default=FALSE, |
| 120 |
help="Generate new profile.") |
| 121 |
parser$add_argument( |
| 122 |
"-c", "--calculate", action="store_true", default=FALSE, |
| 123 |
help="Calculate error rate on existing profile.") |
| 124 |
parser$add_argument( |
| 125 |
"-p", "--profile", help="The source ART error profile.") |
| 126 |
parser$add_argument( |
| 127 |
"-e", "--error-rate", type="double", action="store_true", default=FALSE, |
| 128 |
help="Calculate error rate of existing profile.") |
| 129 |
parser$add_argument( |
| 130 |
"-o", "--output", default=NULL, |
| 131 |
help="New profile file.") |
| 132 |
parser$add_argument( |
| 133 |
"-t", "--table", default=NULL, |
| 134 |
help="File where data will be written in tab-delimited format.") |
| 135 |
args <- parser$parse_args() |
| 136 |
|
| 137 |
if (args$generate) { |
| 138 |
# for some reason, the acutal error rate in simulated reads is about twice |
| 139 |
# what is expected based on the empirical distribution of quality scores |
| 140 |
error.rate <- as.numeric(args$error_rate) / 2 |
| 141 |
adjust.profile(args$profile, args$output, error.rate) |
| 142 |
actual.error <- profile.error(args$output) |
| 143 |
print(paste("The acutal error rate of the new profile is", actual.error)) |
| 144 |
} else if (args$error) { |
| 145 |
print(paste("The acutal error rate of the profile is", round(profile.error(args$profile), 4))) |
| 146 |
} |
| 147 |
|
| 148 |
if (args$table) { |
| 149 |
if (!file.exists(args$table)) { |
| 150 |
write( |
| 151 |
paste("Profile", "RequestedErrorRate", "ActualErrorRate", sep="\t"), |
| 152 |
file=args$table) |
| 153 |
} |
| 154 |
write( |
| 155 |
paste(basename(args$profile), error.rate, actual.error, sep="\t"), |
| 156 |
file=args$table, append=TRUE) |
| 157 |
} |