R/duplicateTxps.R

In tximeta: Transcript Quantification Import with Automatic Metadata

makeDuplicateTxpsList <- function(txomeInfo) {
  dup.list.id <- paste0("dups-",substr(txomeInfo$sha256,1,32))
  bfcloc <- getBFCLoc()
  bfc <- BiocFileCache(bfcloc)
  q <- bfcquery(bfc, dup.list.id)
  if (bfccount(q) == 1) {
    loadpath <- bfcrpath(bfc, dup.list.id)
    dup.list <- readRDS(loadpath)
    stopifnot(is(dup.list, "CharacterList"))
  } else {
    # download the cDNA from remote source (or load from cache)
    dna <- retrieveCDNA(txomeInfo, quiet=TRUE)
    if (txomeInfo$source == "Ensembl") {
      testTxp <- names(dna)[1]
      if (grepl("ENST|ENSMUST",testTxp)) {
        # for human and mouse, need to split on version number
        # (GTF has no version in txname, while FASTA has versions)
        names(dna) <- sub("\\..*", "", names(dna))
      } else {
        names(dna) <- sub(" .*", "", names(dna))
      }
    } else if (txomeInfo$source == "GENCODE") {
      names(dna) <- sub("\\|.*", "", names(dna))
    }
    # which are duplicated based on DNA sequence
    dups <- duplicated(dna)
    if (sum(dups) == 0) {
      dup.list <- NULL
    } else {
      # index of all seqs that participate in a dup
      all.dups <- dna %in% dna[dups]
      nms.all.dups <- names(dna)[all.dups]
      seq.all.dups <- as.character(dna[all.dups])
      dup.list <- split(nms.all.dups, seq.all.dups)
      names(dup.list) <- NULL
      dup.list <- CharacterList(dup.list)
    }
    # save it to BFC so as not to repeat the above operation
    savepath <- bfcnew(bfc, dup.list.id, ext=".rds")
    saveRDS(dup.list, file=savepath)
  }
  dup.list
}

makeDuplicateTxpsTable <- function(missing.txps, dup.list, txp.nms) {
  # dup.list is a list of the duplicate txps, according to the FASTA
  all.dups <- unlist(dup.list)

  # we want to try to fix those duplicate txps that are
  # in rownames of the assays but not in `txps`
  dups.to.fix <- intersect(all.dups, missing.txps)
  dups.to.fix.list <- LogicalList(split(all.dups %in% dups.to.fix, rep(seq_along(dup.list), lengths(dup.list))))
  dup.list <- dup.list[any(dups.to.fix.list)]
  dups.to.fix.list <- dups.to.fix.list[any(dups.to.fix.list)]
  names(dups.to.fix.list) <- NULL

  # if we have a set of duplicates where > 1 is in the rownames of assays,
  # this is unexpected (Salmon should have combined these)
  # and we pass on fixing this
  more.than.one <- sapply(dups.to.fix.list, sum) > 1
  dup.list <- dup.list[!more.than.one]
  dups.to.fix.list <- dups.to.fix.list[!more.than.one]

  # now we can name the duplicate sets by the missing txp
  names(dup.list) <- dup.list[dups.to.fix.list]
  all.dups <- unlist(dup.list)
  
  # is there an alternative in `txps`?
  dups.with.alt <- intersect(all.dups, txp.nms)
  dups.with.alt.list <- LogicalList(split(all.dups %in% dups.with.alt, rep(seq_along(dup.list), lengths(dup.list))))
  dup.list <- dup.list[any(dups.with.alt.list)]
  dups.with.alt.list <- dups.with.alt.list[any(dups.with.alt.list)]

  # pull out the alts
  names(dups.with.alt.list) <- names(dup.list)
  alts <- dup.list[dups.with.alt.list]

  # no reason to prefer one since they are all in `txps`
  alts <- sapply(alts, `[`, 1)
  stopifnot(all(alts %in% txp.nms))

  # return a table for the txp name swap
  dup.table <- data.frame(dups.to.fix=names(alts),
                          alts=unname(alts),
                          stringsAsFactors=FALSE)
  dup.table
}