Skip to content
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 8 additions & 8 deletions R/madc2vcf_all.R
Original file line number Diff line number Diff line change
Expand Up @@ -75,7 +75,7 @@ madc2vcf_all <- function(madc = NULL,
"out_vcf= ", out_vcf, ', ',
"verbose= ", verbose,')">')

if(!is.null(madc)) report <- read.csv(madc) else stop("Please provide a MADC file")
if(!is.null(madc)) report <- read.csv(madc, check.names = FALSE) else stop("Please provide a MADC file")
if(!is.null(botloci_file)) botloci <- read.csv(botloci_file, header = F) else stop("Please provide a botloci file")
if(!is.null(hap_seq_file)) hap_seq <- read.table(hap_seq_file, header = F) else hap_seq <- NULL

Expand Down Expand Up @@ -142,12 +142,12 @@ loop_though_dartag_report <- function(report, botloci, hap_seq, n.cores=1, align
new.file <- data.frame("AlleleID" = report[,1],
"Chromosome" = NA, "SNP_position_in_Genome" = NA,
"Ref" = NA, "Alt" =NA,
"CloneID" = report[,2], report[,3:ncol(report)])
"CloneID" = report[,2], report[,3:ncol(report)], check.names = FALSE)

by_cloneID <- split.data.frame(new.file, new.file$CloneID)

clust <- makeCluster(n.cores)
#clusterExport(clust, c("hap_seq","add_ref_alt"))
#clusterExport(clust, c("hap_seq","add_ref_alt", "nsamples"))
add_ref_alt_results <- parLapply(clust, by_cloneID, function(x) add_ref_alt(x, hap_seq, nsamples, verbose = verbose))
stopCluster(clust)

Expand All @@ -169,7 +169,6 @@ loop_though_dartag_report <- function(report, botloci, hap_seq, n.cores=1, align

clust <- makeCluster(n.cores)
#clusterExport(clust, c("botloci", "compare", "nucleotideSubstitutionMatrix", "pairwiseAlignment", "DNAString", "reverseComplement"))
#clusterExport(clust, c("botloci"))
compare_results <- parLapply(clust, updated_by_cloneID, function(x) compare(x, botloci, alignment_score_thr))
stopCluster(clust)

Expand Down Expand Up @@ -209,7 +208,6 @@ loop_though_dartag_report <- function(report, botloci, hap_seq, n.cores=1, align
#'
#' @noRd
add_ref_alt <- function(one_tag, hap_seq, nsamples, verbose = TRUE) {

# Add ref and alt
cloneID <- one_tag$CloneID[1]
ref <- paste0(cloneID, "|Ref_0001")
Expand Down Expand Up @@ -281,13 +279,14 @@ add_ref_alt <- function(one_tag, hap_seq, nsamples, verbose = TRUE) {
#'
#' @noRd
compare <- function(one_tag, botloci, alignment_score_thr = 40){
#one_tag <- updated_by_cloneID[[2]]
cloneID <- one_tag$CloneID[1]
isBotLoci <- cloneID %in% botloci[,1]
# If marker is present in the botloc list, get the reverse complement sequence
if(isBotLoci) one_tag$AlleleSequence <- sapply(one_tag$AlleleSequence, function(x) as.character(reverseComplement(DNAString(x))))

chr <- sapply(strsplit(cloneID, "_"), function(x) x[-length(x)])
if(length(chr > 1)) chr <- paste(chr, collapse = "_")

# Target SNP at the position in the ID
ref_seq <- one_tag$AlleleSequence[grep("Ref_0001$",one_tag$AlleleID)]
alt_seq <- one_tag$AlleleSequence[grep("Alt_0002$",one_tag$AlleleID)]
Expand All @@ -305,9 +304,9 @@ compare <- function(one_tag, botloci, alignment_score_thr = 40){
alt_base <- substring(alt_seq, align@subject@mismatch@unlistData, align@subject@mismatch@unlistData)

# If target alternative have N, discard whole tag
if(all(alt_base %in% c("A", "T", "C", "G"))) {
# Always only one polymorphism, if there are more than one, not sure which is the target
if(all(alt_base %in% c("A", "T", "C", "G")) & length(align@pattern@mismatch@unlistData) == 1) {

# Update with new info - always only one polymorphism, if there are more than one, not sure which is the target
update_tag <- one_tag[grep("Ref_0001$",one_tag$AlleleID),]
update_tag[,2:5] <- c(chr, pos_target, ref_base, alt_base)
update_tag_temp <- one_tag[grep("Alt_0002$",one_tag$AlleleID),]
Expand Down Expand Up @@ -365,6 +364,7 @@ compare <- function(one_tag, botloci, alignment_score_thr = 40){
return(list(update_tag = NULL,
rm_score = cloneID,
rm_N = NULL))

}
}

Expand Down