R/21.get_UTR3eSet.R

In jianhong/InPAS: Identify Novel Alternative PolyAdenylation Sites (PAS) from RNA-seq data

#' prepare 3' UTR coverage data for usage test
#'
#' generate a UTR3eSet object with PDUI information for statistic tests
#'
#' @param sqlite_db A path to the SQLite database for InPAS, i.e. the output of
#'   [setup_sqlitedb()].
#' @param normalize A character(1) vector, spcifying the normalization method.
#'   It can be "none", "quantiles", "quantiles.robust", "mean", or "median"
#' @param ... parameter can be passed into
#'   [preprocessCore::normalize.quantiles.robust()]
#' @param singleSample A logical(1) vector, indicating whether data is prepared
#'   for analysis in a singleSample mode? Default, FALSE
#'
#' @return An object of [UTR3eSet-class] which contains following elements:
#'   usage: an [GenomicRanges::GRanges-class] object with CP sites info. PDUI: a
#'   matrix of PDUI PDUI.log2: log2 transformed PDUI matrix short: a matrix of
#'   usage of short form long: a matrix of usage of long form if singleSample is
#'   TRUE, one more element, signals, will be included.
#' @export
#' @import GenomicRanges
#' @importFrom preprocessCore colSummarizeAvg colSummarizeMedian
#'   normalize.quantiles normalize.quantiles.robust
#' @author Jianhong Ou, Haibo Liu
#'
#' @examples
#' if (interactive()) {
#'   library(BSgenome.Mmusculus.UCSC.mm10)
#'   library(TxDb.Mmusculus.UCSC.mm10.knownGene)
#'   genome <- BSgenome.Mmusculus.UCSC.mm10
#'   TxDb <- TxDb.Mmusculus.UCSC.mm10.knownGene
#'
#'   ## load UTR3 annotation and convert it into a GRangesList
#'   data(utr3.mm10)
#'   utr3 <- split(utr3.mm10, seqnames(utr3.mm10), drop = TRUE)
#'
#'   bedgraphs <- system.file("extdata", c(
#'     "Baf3.extract.bedgraph",
#'     "UM15.extract.bedgraph"
#'   ),
#'   package = "InPAS"
#'   )
#'   tags <- c("Baf3", "UM15")
#'   metadata <- data.frame(
#'     tag = tags,
#'     condition = c("Baf3", "UM15"),
#'     bedgraph_file = bedgraphs
#'   )
#'   outdir <- tempdir()
#'   write.table(metadata,
#'     file = file.path(outdir, "metadata.txt"),
#'     sep = "\t", quote = FALSE, row.names = FALSE
#'   )
#'
#'   sqlite_db <- setup_sqlitedb(metadata = file.path(
#'     outdir,
#'     "metadata.txt"
#'   ), outdir)
#'   addLockName(filename = tempfile())
#'   coverage <- list()
#'   for (i in seq_along(bedgraphs)) {
#'     coverage[[tags[i]]] <- get_ssRleCov(
#'       bedgraph = bedgraphs[i],
#'       tag = tags[i],
#'       genome = genome,
#'       sqlite_db = sqlite_db,
#'       outdir = outdir,
#'       chr2exclude = "chrM"
#'     )
#'   }
#'
#'   data4CPsSearch <- setup_CPsSearch(sqlite_db,
#'     genome,
#'     chr.utr3 = utr3[["chr6"]],
#'     seqname = "chr6",
#'     background = "10K",
#'     TxDb = TxDb,
#'     hugeData = TRUE,
#'     outdir = outdir,
#'     minZ = 2,
#'     cutStart = 10,
#'     MINSIZE = 10,
#'     coverage_threshold = 5
#'   )
#'   ## polyA_PWM
#'   load(system.file("extdata", "polyA.rda", package = "InPAS"))
#'
#'   ## load the Naive Bayes classifier model from the cleanUpdTSeq package
#'   library(cleanUpdTSeq)
#'   data(classifier)
#'
#'   CPs <- search_CPs(
#'     seqname = "chr6",
#'     sqlite_db = sqlite_db,
#'     genome = genome,
#'     MINSIZE = 10,
#'     window_size = 100,
#'     search_point_START = 50,
#'     search_point_END = NA,
#'     cutEnd = 0,
#'     adjust_distal_polyA_end = TRUE,
#'     long_coverage_threshold = 2,
#'     PolyA_PWM = pwm,
#'     classifier = classifier,
#'     classifier_cutoff = 0.8,
#'     shift_range = 100,
#'     step = 5,
#'     outdir = outdir
#'   )
#'   utr3_cds_cov <- get_regionCov(
#'     chr.utr3 = utr3[["chr6"]],
#'     sqlite_db,
#'     outdir,
#'     phmm = FALSE
#'   )
#'   eSet <- get_UTR3eSet(sqlite_db,
#'     normalize = "none",
#'     singleSample = FALSE
#'   )
#'   test_out <- test_dPDUI(
#'     eset = eSet,
#'     method = "fisher.exact",
#'     normalize = "none",
#'     sqlite_db = sqlite_db
#'   )
#' }
get_UTR3eSet <- function(sqlite_db,
                         normalize = c(
                           "none", "quantiles",
                           "quantiles.robust",
                           "mean", "median"
                         ),
                         ...,
                         singleSample = FALSE) {
  if (missing(sqlite_db) || length(sqlite_db) != 1 || !file.exists(sqlite_db)) {
    stop("sqlite_db, a path to the SQLite database is required!")
  }
  if (!is.logical(singleSample) || length(singleSample) != 1) {
    stop("singleSample must be a logical(1) vector")
  }
  db_conn <- dbConnect(drv = RSQLite::SQLite(), dbname = sqlite_db)
  utr3_cov <- dbReadTable(db_conn, "utr3cds_coverage")
  dbDisconnect(db_conn)
  if (nrow(utr3_cov) < 1) {
    stop("Coverage data for UTR3 and last CDSs is not available")
  }

  ## load the coverage object for UTR3 and last CDSs
  UTR3_CDS.cov <- lapply(utr3_cov$coverage_file, function(.x) {
    readRDS(.x)
  })
  UTR3_CDS.cov <- unlist(GRangesList(UTR3_CDS.cov))

  normalize <- match.arg(
    arg = normalize,
    choices = c(
      "none", "quantiles",
      "quantiles.robust",
      "mean", "median"
    )
  )

  if ((!singleSample) && "data2" %in% colnames(mcols(UTR3_CDS.cov))) {
    message("Single sample mode is on")
    singleSample <- TRUE
  }
  UTRusage <- UTR3_CDS.cov %>% plyranges::filter(!is.na(source))
  UTRusage <- split(UTRusage, UTRusage$transcript)
  UTRusage <- UTRusage[sapply(UTRusage, length) == 2]
  UTRusage <- unlist(UTRusage, use.names = FALSE)
  UTRusage.total <- UTRusage[UTRusage$source == "short"]
  UTRusage.long <- UTRusage[UTRusage$source == "long"]
  UTRusage.total <- UTRusage.total[!duplicated(UTRusage.total$transcript)]
  UTRusage.long <- UTRusage.long[match(
    UTRusage.total$transcript,
    UTRusage.long$transcript
  )]
  PDUItable <- UTRusage.total
  PDUItable$data <- NULL
  start(PDUItable)[as.character(strand(PDUItable)) == "-"] <-
    PDUItable$Predicted_Distal_APA[as.character(strand(PDUItable)) == "-"]
  end(PDUItable)[as.character(strand(PDUItable)) == "+"] <-
    PDUItable$Predicted_Distal_APA[as.character(strand(PDUItable)) == "+"]

  UTRusage.total.data <- do.call(rbind, UTRusage.total$data)
  UTRusage.long.data <- do.call(rbind, UTRusage.long$data)
  UTRusage.short.data <- UTRusage.total.data - UTRusage.long.data
  lt0 <- apply(UTRusage.short.data, 1, function(.ele) any(.ele < 0))
  if (any(lt0)) {
    CDSusage <- UTR3_CDS.cov %>%
      plyranges::filter(feature == "CDS" &
        transcript %in% unique(PDUItable$transcript[lt0]))
    CDSusage.data <- do.call(rbind, CDSusage$data)
    rownames(CDSusage.data) <- CDSusage$transcript
    idx <- match(PDUItable$transcript[lt0], CDSusage$transcript)

    ## why this is reasonable?
    UTRusage.short.data[lt0[!is.na(idx)]] <-
      CDSusage.data[idx[!is.na(idx)], ] -
      UTRusage.long.data[lt0[!is.na(idx)]]
    UTRusage.short.data[UTRusage.short.data < 0] <- 0
  }

  ## normalization?
  normalize.foo <- function(exprs, FUN) {
    avgs <- FUN(exprs)$Estimates
    scaling.factors <- avgs / avgs[1]
    scaling.factors <- matrix(rep(scaling.factors, nrow(exprs)),
      ncol = length(scaling.factors), byrow = TRUE
    )
    exprs <- exprs / scaling.factors
  }
  normalize.mean <- function(exprs) {
    normalize.foo(exprs, colSummarizeAvg)
  }
  normalize.median <- function(exprs) {
    normalize.foo(exprs, colSummarizeMedian)
  }
  if (normalize != "none") {
    UTRusage.long.short.data <- rbind(
      UTRusage.long.data,
      UTRusage.short.data
    )
    UTRusage.long.short.data <-
      switch(normalize,
        quantile = normalize.quantiles(UTRusage.long.short.data),
        quantile.robust =
          normalize.quantiles.robust(UTRusage.long.short.data, ...),
        mean = normalize.mean(UTRusage.long.short.data),
        median = normalize.median(UTRusage.long.short.data),
        UTRusage.long.short.data
      )
    ## suppose the output should be same order
    UTRusage.long.data <-
      UTRusage.long.short.data[1:nrow(UTRusage.long.data), , drop = FALSE]
    UTRusage.short.data <-
      UTRusage.long.short.data[-(1:nrow(UTRusage.long.data)), ,
        drop = FALSE
      ]
  }

  UTRusage.PDUI <-
    UTRusage.long.data / (UTRusage.long.data + UTRusage.short.data)
  UTRusage.PDUI.log2 <- log2(UTRusage.PDUI + .Machine$double.xmin)
  rownames(UTRusage.PDUI) <-
    rownames(UTRusage.PDUI.log2) <-
    rownames(UTRusage.long.data) <-
    rownames(UTRusage.short.data) <- PDUItable$transcript
  PDUItable$source <- NULL
  if (singleSample) {
    signals.short <- UTRusage.total$data2
    signals.long <- UTRusage.long$data2

    ## ??
    cut50 <- function(x, y) {
      z <- c(x, y)
      if (length(z) > 50) {
        xat <- floor(50 * length(x) / length(z))
        z <- tapply(z, cut(1:length(z), 50), mean)
      } else {
        xat <- length(x)
      }
      c(xat, z)
    }
    signals <- mapply(cut50, signals.short,
      signals.long,
      SIMPLIFY = FALSE
    )
    names(signals) <- UTRusage.total$transcript
    new("UTR3eSet",
      usage = PDUItable,
      PDUI = UTRusage.PDUI,
      PDUI.log2 = UTRusage.PDUI.log2,
      short = UTRusage.short.data,
      long = UTRusage.long.data,
      signals = signals
    )
  } else {
    new("UTR3eSet",
      usage = PDUItable,
      PDUI = UTRusage.PDUI,
      PDUI.log2 = UTRusage.PDUI.log2,
      short = UTRusage.short.data,
      long = UTRusage.long.data
    )
  }
}