#' @export
InitDR2S.DR2Sconf <- function(config, createOutdir = TRUE) {
DR2S_$new(config, createOutdir)
}
#' @export
cache.DR2S <- function(x, outname, ...) {
if (missing(outname)) {
outname <- dot(c("DR2S", x$getLrdType(), x$getLrdMapper(), "rds"))
}
x$cache(outname = outname)
invisible(x)
}
#' @export
clear.DR2S <- function(x, ...) {
x$clear()
invisible(x)
}
# Class: DR2S -------------------------------------------------------------
#' R6 Class \code{"DR2S"}
#'
#' @description An object representing a DR2S analysis.
#'
DR2S_ <- R6::R6Class(
classname = "DR2S",
public = list(
##
## Public fields
##
#' @field mapInit <MapList>; the initial mapping of long reads to
#' \code{reference}.
mapInit = list(),
#' @field lrpartition <PartList>; the partitioning of full-length
#' mapped long reads into different haplotypes.
lrpartition = list(),
#' @field mapIter <MapList>; the mapping of A and B reads to
#' consensus sequences produced in the previous step.
mapIter = list(),
#' @field srpartition <PartList>; the partitioning of mapped short reads
#' into different haplotypes.
srpartition = list(),
#' @field mapFinal <MapList>; the mapping of A, B, and short reads
#' to consensus sequences produced in the previous step.
mapFinal = list(),
#' @field remap <MapList>; the mapping of A, B, and short reads
#' to consensus sequences produced in the previous step.
remap = list(),
#' @field consensus <ConsList>; final consensus sequences for A and
#' B.
consensus = list(),
##
## Public functions
##
#' @description Create a new DR2S object.
#' @param conf A DR2S configuration provided either by [createDR2SConf] or
#' read from file by [readDR2SConf].
#' @param createOutdir Set up the output directory.
#' @return A new DR2S object.
initialize = function(conf, createOutdir = TRUE) {
private$conf = initialiseDR2S(conf, createOutdir = createOutdir)
private$runstats = NULL
private$homozygous = FALSE
if (is.null(conf$reference) ||
conf$reference == conf$locus) {
## fetch the generic reference for <locus>
private$conf$reference = .normaliseLocus(conf$locus)
private$runstats$refPath = .generateReferenceSequence(
locus = private$conf$locus,
allele = NULL,
outdir = private$conf$outdir,
utrLength = private$conf$utrLength,
dirtag = "mapInit")
} else if (file.exists(refPath <- conf$reference)) {
## use the user-provided reference for <locus>
private$conf$extref = normalizePath(refPath, mustWork = TRUE)
private$conf$reference = sub("\\.fa(s|sta)?$", "", basename(refPath))
private$runstats$refPath = file.path("mapInit", basename(refPath))
cPath <- .dirCreateIfNotExists(normalizePath(
file.path(private$conf$outdir, "mapInit"), mustWork = FALSE))
file.copy(refPath, file.path(cPath, basename(refPath)))
} else {
## fetch the allele-specific reference for <locus>
private$conf$reference = .expandAllele(conf$reference, conf$locus)
private$runstats$refPath = .generateReferenceSequence(
locus = private$conf$locus,
allele = private$conf$reference,
outdir = private$conf$outdir,
utrLength = private$conf$utrLength,
dirtag = "mapInit")
}
.confLog(outdir = private$conf$outdir, logName = "info")
private$srStatus = self$checkGetShortreads()
private$lrStatus = self$checkGetLongreads()
writeDR2SConf(self)
flog.info("Creating DR2S Object", name = "info")
},
#' @description Store a DR2S object as an rds file.
#' @param outname (optional) filename of the stored object.
#' @return Returns invisibly the DR2S object.
cache = function(outname) {
if (missing(outname)) {
outname <- dot(c("DR2S", self$getLrdType(), self$getLrdMapper(), "rds"))
}
path <- file.path(self$getOutdir(), outname)
flog.info("Caching <%s>", path, name = "info")
saveRDS(self$clone(deep = TRUE), file = path, compress = FALSE)
invisible(self)
},
#' @description Clear all the files pertaining to a DR2S mapping process
#' and set up the whole process from scratch.
#' @return Returns invisibly the DR2S object.
clear = function() {
unlink(self$getOutdir(), recursive = TRUE)
.dirCreateIfNotExists(file.path(self$getOutdir()))
if (!is.null(private$conf$extref) && file.exists(refPath <- private$conf$extref)) {
## use the user-provided reference for <locus>
private$conf$reference = sub("\\.fa(s|sta)?$", "", basename(refPath))
private$runstats$refPath = file.path("mapInit", basename(refPath))
cPath <- .dirCreateIfNotExists(normalizePath(
file.path(private$conf$outdir, "mapInit"), mustWork = FALSE))
file.copy(refPath, file.path(cPath, basename(refPath)))
} else if (self$getReference() == self$getLocus()) {
## fetch the generic reference for <locus>
private$runstats$refPath = .generateReferenceSequence(
locus = self$getLocus(),
allele = NULL,
outdir = self$getOutdir(),
utrLength = self$getConfig("utrLength"),
dirtag = "mapInit")
} else {
## fetch the allele-specific reference for <locus>
outdir <- .dirCreateIfNotExists(normalizePath(
self$absPath("mapInit"), mustWork = FALSE))
private$runstats$refPath = .generateReferenceSequence(
locus = self$getLocus(),
allele = self$getReference(),
outdir = self$getOutdir(),
utrLength = self$getConfig("utrLength"),
dirtag = "mapInit")
}
.confLog(outdir = self$getOutdir(), logName = "info")
private$srStatus = self$checkGetShortreads()
private$lrStatus = self$checkGetLongreads()
writeDR2SConf(self)
invisible(self)
},
#' @description remove all bam and bai files.
#' @return Returns invisibly the DR2S object.
cleanup = function() {
unlink(self$absPath(bampath(self$mapInit)))
unlink(self$absPath(bampath(self$mapInit)) %<<% ".bai")
foreach(hpt = self$getHapTypes()) %do% {
unlink(self$absPath(hpt), recursive = TRUE)
}
unlink(self$absPath("final"), recursive = TRUE)
invisible(self)
},
#' @description Print a DR2S object prettily
#' @return Returns invisibly the DR2S object.
print = function() {
fmt0 <- "DR2S driver for sample <%s> locus <%s>\n"
cat(sprintf(fmt0, self$getSampleId(), self$getLocus()))
fmt1 <- "Reference: <%s>\n" %<<%
"Longreads: <%s> Shortreads: <%s>\n" %<<%
"Mapper: <%s>\nDatadir: <%s>\nOutdir: <%s>\n"
cat(sprintf(fmt1,
self$getReference(), self$getLrdType(),
self$getSrdType() %||% "", self$getLrdMapper(),
self$getDatadir(), self$getOutdir()
))
invisible(self)
},
#' @description Execute a given step in a DR2S pipeline.
#' @param step A step in the DR2S pipeline
#' @return Called for side effect.
run_ = function(step) {
switch(
step,
clear = self$clear(),
cache = self$cache(),
partitionLongreads = {
self$runPartitionLongreads()
self$runSplitLongreadsByHaplotype()
self$runExtractPartitionedLongreads()
},
partitionShortreads = {
self$runPartitionShortreads()
},
mapInit = self$runMapInit(),
mapIter = self$runMapIter(),
mapFinal = self$runMapFinal(),
report = DR2S::report(self),
stop("<", step, "> is not a valid step in the mapping pipeline")
)
},
##
## Config getters and setters ####
##
getConfig = function(name = NULL) {
if (is.null(name))
private$conf
else
private$conf[[name]]
},
##
setConfig = function(name, value) {
private$conf[name] = value
invisible(self)
},
##
getDatadir = function() {
self$getConfig("datadir")
},
##
getOutdir = function() {
self$getConfig("outdir")
},
##
getMicrosatellite = function() {
self$getOpts("mapInit")$microsatellite
},
##
getForceMapping = function() {
self$getOpts("mapInit")$forceMapping
},
##
getThreshold = function() {
self$getOpts("partitionLongreads")$threshold
},
##
getDistAlleles = function() {
self$getOpts("partitionLongreads")$distAlleles
},
##
getIterations = function() {
self$getOpts("mapIter")$iterations
},
##
getLrdMapper = function() {
self$getConfig("longreads")$mapper
},
##
getLrdType = function() {
self$getConfig("longreads")$type
},
##
getLrdDir = function() {
lrddir <- self$getConfig("longreads")$dir
if (is.null(lrddir)) {
self$getDatadir()
} else {
file.path(self$getDatadir(), lrddir)
}
},
##
getLongreads = function() {
lrdfile <- self$getConfig("longreads")$file
if (!is.null(lrdfile)) {
readpath <- file.path(self$getDatadir(), lrdfile)
names(readpath) <- lrdfile
} else {
lrddir <- self$getLrdDir()
readpath <- findReads(datadir = lrddir,
sampleId = self$getSampleId(),
locus = self$getLocus())
names(readpath) <- .cropPath(self$getDatadir(), readpath)
}
if (is.null(readpath) || length(readpath) == 0 || !file.exists(readpath)) {
flog.error("No reads available for readtype <%s>", self$getLrdType(), name = "info")
stop("No reads available for readtype <", self$getLrdType(), ">")
}
readpath
},
##
getSrdMapper = function() {
self$getConfig("shortreads")$mapper
},
##
getSrdType = function() {
self$getConfig("shortreads")$type
},
##
getSrdDir = function() {
srddir <- self$getConfig("shortreads")$dir
if (!is.null(sdir <- self$mapInit$meta$SR1$readpath)) {
if (!any(startsWith(sdir, srddir))) {
self$absPath(sdir)
}
}
if (!is.null(srddir)) {
file.path(self$getDatadir(), srddir)
} else {
NULL
}
},
##
getShortreads = function() {
filtered <- self$getConfig("trimmedShortreads")
if (!is.null(filtered)) {
return(self$absPath(filtered))
}
srddir <- self$getSrdDir()
if (is.null(srddir)) {
return(NULL)
}
readpath <- findReads(srddir, self$getSampleId(), self$getLocus())
names(readpath) <- .cropPath(self$getDatadir(), readpath)
if (is.null(readpath) || length(readpath) == 0 || !file.exists(readpath)) {
flog.error("No reads available for readtype <%s>", self$getSrdType(), name = "info")
stop("No reads available for readtype <", self$getSrdType(), ">")
}
readpath
},
##
getPipeline = function() {
switch(self$getConfig("pipeline"), LR = LR_PIPELINE(), SR = SR_PIPELINE())
},
##
getOpts = function(name = NULL) {
opts <- self$getConfig("opts")
if (is.list(opts))
opts[[name]]
else
opts
},
##
getSampleId = function() {
self$getConfig("sampleId")
},
##
getLocus = function() {
self$getConfig("locus")
},
##
getReference = function() {
self$getConfig("reference")
},
##
getPlatform = function() {
self$getDetails("platform") %||%
self$getConfig("longreads")$platform %||%
self$getConfig("longreads")$type
},
##
getFormat = function() {
self$getConfig("format")
},
##
setFormat = function(format) {
assert_that(is.character(format) ,format %in% c("yaml", "json"))
self$setConfig("format", format)
invisible(self)
},
##
getDetails = function(name = NULL) {
if (is.null(name)) {
self$getConfig("details")
} else {
assert_that(is.character(name))
self$getConfig("details")[[name]]
}
},
##
setDetails = function(details) {
assert_that(is.character(details))
self$setConfig("details", details)
invisible(self)
},
##
getSampleDetails = function() {
details <- semicolon(
vapply(seq_along(self$getDetails()),
function(item) paste(names(
self$getDetails()[item]),
underscore(litQuote(self$getDetails()[item])),
sep = "="), FUN.VALUE = character(1)))
sr <- self$hasShortreads()
lr <- self$hasLongreads()
paste("locus=" %<<% litQuote(self$getLocus()),
"ref=" %<<% litQuote(self$getReference()),
details,
"short_read_data=" %<<% litQuote(ifelse(sr, "yes", "no")),
"short_read_type=" %<<% litQuote(ifelse(sr, self$getSrdType(), "")),
"long_read_data=" %<<% litQuote(ifelse(lr, "yes", "no")),
"long_read_type=" %<<% litQuote(ifelse(lr, self$getLrdType(), "")),
"software=\"DR2S\"",
"version=" %<<% litQuote(packageVersion("DR2S")),
sep = ";")
},
##
## Runstats getters and setters ####
##
getStats = function(name = NULL) {
if (is.null(name))
private$runstats
else
private$runstats[[name]]
},
##
setStats = function(name, value) {
private$runstats[name] = value
invisible(self)
},
##
getLimits = function() {
self$getStats("partitionLongreads")$limits
},
##
setLimits = function(lmts) {
# stopifnot(is.null(lmts) || all(lmts >=0 && lmts <=1))
x <- list()
private$runstats$partitionLongreads$limits = lmts
invisible(self)
},
##
getHapTypes = function() {
self$getStats("partitionLongreads")$haplotypes
},
##
setHapTypes = function(x) {
## Todo: add test for data consistency!
private$runstats$partitionLongreads$haplotypes = x
invisible(self)
},
##
getRefPath = function() {
self$absPath(self$getStats("refPath"))
},
##
setRefPath = function(refPath) {
self$setStats("refPath", self$relPath(refPath))
},
##
## Report Status ####
##
getHomozygous = function() {
private$homozygous
},
##
setHomozygous = function(homozygous) {
assert_that(is.logical(homozygous))
private$homozygous <- homozygous
invisible(self)
},
##
## Internal getters/setters ####
##
getRefSeq = function() {
if (!is.null(self$getRefPath())) {
Biostrings::readDNAStringSet(self$getRefPath())
} else {
if (startsWith(self$getLocus, "KIR")) {
.ipdKir()$getClosestComplete(self$getReference(),
self$getLocus())
} else {
.ipdHla()$getClosestComplete(self$getReference(),
self$getLocus())
}
}
},
##
getPileup = function(useSR = FALSE) {
if (useSR)
meta(self$mapInit, "SR2")$pileup
else
self$mapInit$pileup
},
##
getSnpMatrix = function() {
self$lrpartition$mat
},
##
getPartition = function() {
self$lrpartition$prt
},
##
getHapList = function(group) {
if (missing(group)) {
self$lrpartition$hpl
} else {
self$lrpartition$hpl[[group]]
}
},
##
getLatestRefPath = function() {
##
if (self$hasMapFinal() && self$hasShortreads()) {
return(vapply(self$mapFinal$SR, function(x) self$absPath(conspath(x)),
FUN.VALUE = character(1)))
}
##
if (self$hasMapFinal() && self$hasLongreads()) {
return(vapply(self$mapFinal$LR, function(x) self$absPath(conspath(x)),
FUN.VALUE = character(1)))
}
##
if (self$hasMapIter()) {
latest <- self$mapIter[[
max(names(self$mapIter))]][self$getHapTypes()]
return(vapply(latest, function(x) self$absPath(conspath(x)),
FUN.VALUE = character(1)))
}
##
return(self$getRefPath())
},
##
getLatestRef = function() {
filepath <- self$getLatestRefPath()
Biostrings::readDNAStringSet(filepath, use.names = TRUE)
},
##
getMapTag = function(iter = "init", group = NULL, ref = "LR") {
if (is.numeric(iter)) {
group <- match.arg(group, self$getHapTypes())
return(tag(self$mapIter[[as.character(iter)]][[group]]))
}
if (iter == "init") {
ref <- match.arg(ref, c("LR", "SR"))
if (ref == "SR") {
return(tag(meta(self$mapInit, "SR1")))
}
return(tag(self$mapInit))
}
if (iter == "final") {
group <- match.arg(group, self$getHapTypes())
ref <- match.arg(ref, c("LR", "SR"))
return(tag(self$mapFinal[[ref]][[group]]))
}
if (iter == "remap") {
group <- match.arg(group, self$getHapTypes())
ref <- match.arg(ref, c("LR", "SR"))
return(tag(self$remap[[ref]][[group]]))
}
},
##
getGroupPileup = function(group = NULL, ref = NULL, iteration = 0) {
group <- match.arg(group, self$getHapTypes())
ref <- match.arg(ref, c("LR", "SR"))
if (is.numeric(iteration)) {
self$mapIter[[as.character(iteration)]][[group]]$pileup
} else if (iteration == "remap") {
self$remap[[ref]][[group]]$pileup
} else {
self$mapFinal[[ref]][[group]]$pileup
}
},
##
getLrdMapFun = function() {
match.fun("run" %<<% self$getLrdMapper())
},
##
getSrdMapFun = function() {
match.fun("run" %<<% self$getSrdMapper())
},
## Get the absolut path
absPath = function(filename) {
assert_that(is.character(filename))
## set relpath as name attribute
names(filename) <- filename
vapply(filename, function(x) {
normalizePath(
file.path(self$getOutdir(), x),
mustWork = FALSE)
}, FUN.VALUE = character(1), USE.NAMES = TRUE)
},
## Get the relative path
relPath = function(filepath) {
## TODO add assertives
if (all(startsWith(filepath, self$getOutdir()))) {
.cropPath(base = self$getOutdir(), path = filepath)
} else if (all(startsWith(filepath, self$getDatadir()))) {
.cropPath(base = self$getDatadir(), path = filepath)
} else {
filepath
}
},
##
## Predicate methods ####
##
hasMapInit = function() {
tryCatch(
is(self$mapInit, "MapList"),
error = function(e)
FALSE
)
},
##
hasLrdPartition = function() {
tryCatch(
is(self$lrpartition$prt, "HapPart"),
error = function(e)
FALSE
)
},
##
hasHapList = function() {
tryCatch(
is(self$lrpartition$hpl, "HapList"),
error = function(e)
FALSE
)
},
##
hasMapIter = function() {
tryCatch(
all(vapply(self$mapIter[[self$getIterations() + 1]],
is, "MapList", FUN.VALUE = logical(1))),
error = function(e)
FALSE
)
},
##
hasMapFinal = function() {
tryCatch(
all(vapply(self$mapFinal$LR, is, "MapList", FUN.VALUE = logical(1))),
error = function(e)
FALSE
)
},
##
checkGetShortreads = function() {
sr <- try(self$getShortreads(), silent = TRUE)
!(is(sr, "try-error") || is.null(sr))
},
##
hasShortreads = function() {
private$srStatus
},
##
checkGetLongreads = function() {
lr <- try(self$getLongreads(), silent = TRUE)
!(is(lr, "try-error") || is.null(lr))
},
##
hasLongreads = function() {
private$lrStatus
},
##
## Plotting methods ####
##
##
plotCoverage = function(threshold,
range = NULL,
thin = 0.1,
width = 1,
label = "",
drop.indels = FALSE,
readtype = "LR") {
tag <- self$getMapTag(ref = readtype)
if (missing(threshold)) {
threshold <- self$getThreshold()
}
pileup <- switch(
readtype,
LR = self$getPileup(useSR = FALSE),
SR = self$getPileup(useSR = TRUE)
)
plotPileupCoverage(
pileup,
threshold,
range,
thin,
width,
label %|ch|% tag,
drop.indels = drop.indels
)
},
##
plotGroupCoverage = function(group = NULL,
ref = NULL,
iteration = "init",
threshold = NULL,
range = NULL,
thin = 0.1,
width = 1,
label = "",
drop.indels = FALSE) {
pileup <- self$getGroupPileup(group, ref, iteration)
lbl <- self$getMapTag(iteration, group)
if (is.null(threshold)) {
threshold <- self$getThreshold()
}
plotPileupCoverage(
x = pileup,
threshold = threshold,
range = range,
thin = thin,
width = width,
label = label %|ch|% lbl,
drop.indels = drop.indels
)
},
##
plotBasecallFrequency = function(threshold, label = "",
drop.indels = FALSE) {
tag <- self$getMapTag()
if (missing(threshold)) {
threshold <- self$getThreshold()
}
plotPileupBasecallFrequency(
self$getPileup(),
threshold,
label %|ch|% tag,
drop.indels = drop.indels
)
},
##
plotGroupBasecallFrequency = function(group = NULL,
ref = NULL,
iteration = "init",
threshold = NULL,
label = "",
drop.indels = FALSE) {
pileup <- self$getGroupPileup(group, ref, iteration)
lbl <- self$getMapTag(iteration, group)
if (is.null(threshold)) {
threshold <- self$getThreshold()
}
plotPileupBasecallFrequency(
pileup,
threshold,
label %|ch|% lbl,
drop.indels = drop.indels
)
},
##
plotPartitionHistogram = function(label = "", limits = NULL) {
tag <- self$getMapTag()
if (length(self$getHapTypes()) == 2) {
plotPartitionHistogram(x = self$getPartition(),
label %|ch|% tag,
limits = limits)
} else {
plotPartitionHistogramMulti(x = self$getPartition(),
limits = limits,
label %|ch|% tag)
}
},
##
plotPartitionTree = function() {
plotPartitionTree(x = self$getPartition())
},
##
plotPartitionRadar = function() {
plotRadarPartition(x = self$getPartition())
},
##
## Summary methods ####
##
polymorphicPositions = function(threshold, useSR = FALSE) {
if (missing(threshold)) {
threshold <- self$getThreshold()
}
pileup <- self$getPileup(useSR = useSR)
polymorphicPositions(pileup, threshold)
},
##
plotMapInitSummary = function(thin = 0.2, width = 4, label = "") {
readtypes <- if (self$hasShortreads()) {
c("SR", "LR")
} else {
c("LR")
}
plotlist <- foreach(readtype = readtypes) %do% {
# readtype <- "LR"
tag <- self$getMapTag(ref = readtype)
self$plotCoverage(
thin = thin,
width = width,
label = label %|ch|% tag,
readtype = readtype
)
}
cowplot::plot_grid(plotlist = plotlist, labels = readtypes,
nrow = length(plotlist))
},
##
plotPartitionSummary = function(label = "", limits = NULL) {
tag <- self$getMapTag()
p1 <- self$plotPartitionHistogram(label = label %||% tag,
limits = limits) +
theme(legend.position = "none")
p2 <- self$plotPartitionTree()
p3 <- self$plotPartitionRadar()
if (is.null(p2)) {
cowplot::plot_grid(p1, p3, ncol = 1)
} else {
cowplot::plot_grid(p1, p2, p3, ncol = 1, rel_heights = c(1,2,1))
}
},
##
plotMapIterSummary = function(thin = 0.2, width = 10, iteration = 0,
drop.indels = TRUE) {
hptypes <- self$getHapTypes()
if (iteration == self$getIterations()) {
drop.indels <- FALSE
width <- 10
}
plotlist <- foreach(hptype = hptypes) %do% {
tag <- self$getMapTag(iteration, hptype)
self$plotGroupCoverage(hptype, ref = NULL, iteration = iteration,
threshold = NULL, range = NULL, thin, width,
label = tag, drop.indels = drop.indels)
}
cowplot::plot_grid(plotlist = plotlist, labels = hptypes)
},
##
plotMapFinalSummary = function(readtype, thin = 0.2, width = 10) {
hptypes <- self$getHapTypes()
## hp = "A"
plotlist <- foreach(hp = hptypes) %do% {
tag <- self$getMapTag(iter = "final", group = hp, ref = readtype)
self$plotGroupCoverage(group = hp, ref = readtype,
iteration = "final", threshold = NULL,
range = NULL, thin = thin,
width = width, label = tag)
}
nrow <- if (self$hasShortreads()) 1 else 2
cowplot::plot_grid(plotlist = plotlist, labels = hptypes, nrow = nrow,
hjust = -2.5)
},
##
plotRemapSummary = function(readtype, thin = 0.2, width = 10) {
hptypes <- self$getHapTypes()
## hp = "A"
plotlist <- foreach(hp = hptypes) %do% {
tag <- self$getMapTag(iter = "remap", group = hp, ref = readtype)
self$plotGroupCoverage(group = hp, ref = readtype,
iteration = "remap", threshold = NULL,
range = NULL, thin = thin,
width = width, label = tag)
}
nrow <- if (self$hasShortreads()) 1 else 2
cowplot::plot_grid(plotlist = plotlist, labels = hptypes, nrow = nrow,
hjust = -2.5)
},
##
plotSeqLogo = function(ppos = NULL, pwm = NULL) {
if (is.null(ppos)) {
ppos <- SNP(self$getPartition())
}
if (is.null(pwm)) {
pwm <- lapply(PWM(self$getPartition()), function(pwm) {
pwm[pwm < 0.1] <- 0
pwm
})
}
ggplot() +
ggseqlogo::geom_logo(pwm, method = "bits", seq_type = "dna", stack_width = 0.9) +
scale_x_continuous(labels = ppos, breaks = seq_along(ppos)) +
facet_wrap(~ seq_group, ncol = 1, strip.position = "left") +
ggseqlogo::theme_logo() +
theme(
axis.text.x = element_text(size = 10, angle = 45),
axis.title.y = element_blank(),
axis.text.y = element_blank(),
axis.ticks.y = element_blank(),
strip.text.y = element_text(face = "bold", size = 42, angle = 180)
)
}
),
##
## Private fields ####
##
private = list(
conf = NULL,
runstats = NULL,
srStatus = FALSE,
lrStatus = FALSE,
homozygous = FALSE
)
)
# Helpers -----------------------------------------------------------------
findReads <- function(datadir, sampleId, locus) {
#sampleId <- self$getSampleId()
#locus <- self$getLocus()
locus <- sub("^HLA-", "", toupper(locus))
locus <- sub("^KIR", "", toupper(locus))
filePattern <- sampleId %<<% "_" %<<% locus %<<% ".+" %<<% "fast(q|a)(\\.gz)?$"
readPath <- dir(datadir, pattern = filePattern, full.names = TRUE)
if (length(readPath) == 0) {
filePattern <- sampleId %<<% ".+" %<<% "fast(q|a)(\\.gz)?$"
readPath <- dir(datadir, pattern = filePattern, full.names = TRUE)
flog.info("Could not determine locus in fastq filename. Assuming its only %s in the shortread directory", locus)
}
readPath <- readPath[grep(pattern = "^((?!_trimmed.fastq).)*$", readPath, perl = TRUE)]
if (length(readPath) > 0) {
normalizePath(readPath, mustWork = TRUE)
} else readPath
}
findRef <- function(locus) {
if (startsWith(locus, "HLA")) {
p <- system.file("extdata", "HLAREF.fa", package = "DR2S")
Biostrings::readDNAStringSet(p, format = "fasta")[locus]
} else if (startsWith(locus, "KIR")) {
stop("No generic KIR references available")
} else if (startsWith(locus, "MIC")) {
stop("No generic MIC references available")
} else {
stop("Unknown locus")
}
}
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.