R/spectrum-functions.R

In synapter: Label-free data analysis pipeline for optimal identification and quantitation

#' parse synapter Spectrum.xml files
#'
#' Stupid parser for one of the ugliest pseudo-xml formats I have ever seen.
#'
#' @param file file path
#' @param encoding file encoding, seems to be windows specific
#' @param verbose verbose output?
#'
#' @return a list of length == 4; names: ms1, m2, leRows, heRows
#'  ms1: matrix of ms1 spectra information
#'  ms2: matrix of ms2 spectra information
#'  leRows: environment (key == leID), values == LE_ID row index
#'  heRows: environment (key == leID), values == HE_ID row index
#' @noRd
.readSynapterSpectrumXml <- function(file, verbose=interactive()) {
  stopifnot(file.exists(file))

  ## helper functions
  .extractName <- function(x) {
    gsub("^.*NAME=\\\"([[:alpha:]_]+)\\\".*$", "\\1", x)
  }
  .extractLEID <- function(x) {
    gsub("^.*LE_ID=\\\"([[:digit:]]+)\\\".*$", "\\1", x)
  }
  .extractHEID <- function(x) {
    gsub("^.*HE_ID=\\\"([[:digit:],]+)\\\".*$", "\\1", x)
  }
  .createMsMatrix <- function(x, header, verbose=interactive()) {
    x <- paste0(gsub("[[:space:]]+", ",", trimws(x)), collapse="\n")
    cols <- intersect(header, c("Mass", "Intensity", "LE_ID", "HE_ID", "Z", "RT"))
    as.matrix(read_csv(x, col_names=header,
                       col_types=do.call(cols_only, as.list(setNames(rep.int("d", length(cols)), cols))),
                       trim_ws=TRUE, progress=verbose))
  }

  if (verbose) {
    message("Reading ", file)
  }
  content <- read_lines(file)

  if (verbose) {
    message("Search line numbers for Header information")
  }
  linesField <- grep("<FIELD", content[1L:min(200L, length(content))],
                     fixed=TRUE)
  splitPoint <- which(diff(linesField) > 1L)
  linesFieldMs1 <- linesField[1:splitPoint]
  linesFieldMs2 <- linesField[(splitPoint+1):length(linesField)]

  if (verbose) {
    message("Read Header information")
  }

  header1 <- .extractName(content[linesFieldMs1])
  header2 <- .extractName(content[linesFieldMs2])

  if (verbose) {
    message("Search line numbers for MS Data")
  }
  linesMs <- grep("</?(HE_)?DATA", content)+c(1L, -1L)

  if (verbose) {
    message("Read MS1 Data (", paste(linesMs[1L:2L], collapse=":"), ")")
  }
  ms1 <- .createMsMatrix(content[linesMs[1L]:linesMs[2L]], header1)

  if (verbose) {
    message("Read MS2 Data (", paste(linesMs[3L:4L], collapse=":"), ")")
  }
  ms2 <- .createMsMatrix(content[linesMs[3L]:linesMs[4L]], header2)

  if (verbose) {
    message("Search line numbers for MS1 to MS2 assignment")
  }
  linesAssignment <- grep("</?PRECURSOR_PRODUCT_BIN", content)+c(1L, -1L)

  if (verbose) {
    message("Read Assignment Data (", paste(linesAssignment, collapse=":"), ")")
  }
  range <- linesAssignment[1L]:linesAssignment[2L]

  if (verbose) {
    message("  extract LE_ID information")
  }
  le_ids <- .extractLEID(content[range])

  if (verbose) {
    message("  extract HE_ID information")
  }
  he_ids <- lapply(MSnbase:::utils.ssv2list(
                     .extractHEID(content[range]), sep=","), as.numeric)

  leRows <- new.env(hash=TRUE, parent=emptyenv(), size=length(le_ids))
  heRows <- new.env(hash=TRUE, parent=emptyenv(), size=length(le_ids))

  if (verbose) {
    message("  match LE_ID and HE_ID information")
  }
  le_rows <- findInterval(le_ids, ms1[, "LE_ID"])
  he_rows <- relist(findInterval(unlist(he_ids), ms2[, "HE_ID"]), he_ids)

  for (i in seq(along=le_ids)) {
    assign(le_ids[i], le_rows[[i]], envir=leRows)
    assign(le_ids[i], he_rows[[i]], envir=heRows)
  }

  list(ms1=ms1, ms2=ms2, leRows=leRows, heRows=heRows)
}

#' read spectrum.xml and turn data into MSnbase::Spectrum2 objects
#' @param df corresponding df from the synapter object ({Ident,Quant}PeptideData)
#' @param file filename
#' @param assaydata environment
#' @param storeAll should all spectra stored? or only the needed ones?
#' @param removePrecursor remove precursor ion from fragments
#' @param tolerance tolerance to look for the precursor ion
#' @param verbose verbose output
#' @return modified assaydata
#' @noRd
.spectrumXml2spectra <- function(df, file, storeAll=TRUE, removePrecursor=TRUE,
                                 tolerance=25e-6, verbose=interactive()) {
  xml <- .readSynapterSpectrumXml(file, verbose=verbose)

  peptideinfo <- df[!duplicated(df$precursor.leID), ]

  if (storeAll) {
    uleID <- union(df$precursor.leID, xml$ms1[, "LE_ID"])
  } else {
    uleID <- intersect(df$precursor.leID, xml$ms1[, "LE_ID"])
  }

  if (verbose) {
    message("Convert spectra data.frames to MSnbase::Spectrum2 objects")
    pb <- txtProgressBar(0, length(uleID), style=3)
  }

  sequences <- rep(NA_character_, length(uleID))

  assaydata <- new.env(hash=TRUE, parent=emptyenv(), size=length(uleID))

  for (i in seq(along=uleID)) {
    assign(as.character(uleID[i]),
           .createMs2SpectrumFromSpectrumXml(uleID[i], xml$ms1, xml$ms2,
                                             xml$leRows, xml$heRows,
                                             removePrecursor=removePrecursor,
                                             tolerance=tolerance),
           envir=assaydata)
    if (i <= nrow(df)) {
      ## in the xml files are no sequence information that's why we have to use
      ## information from the df
      sequences[i] <- peptideinfo$peptide.seq[i]
    }

    if (verbose) {
      setTxtProgressBar(pb, i)
    }
  }

  if (verbose) {
    close(pb)
    message("Create MSnExp object")
  }

  process <- new("MSnProcess",
                 processing=paste("Data loaded:", date()),
                 files=file)

  nm <- ls(assaydata)
  pdata <- new("NAnnotatedDataFrame", data=data.frame(sampleNames=1L,
                                                      type="spectrum"))
  fdata <- data.frame(spectrum=1:length(nm),
                      leID=uleID,
                      peptide.seq=sequences,
                      row.names=uleID,
                      stringsAsFactors=FALSE)
  ## reorder according to assaydata
  fdata <- fdata[nm, ]

  fdata <- new("AnnotatedDataFrame", data=fdata)

  msnexp <- new("MSnExp", assayData=assaydata,
                          phenoData=pdata,
                          featureData=fdata,
                          processingData=process)
  ## don't do any validation here, we check each spectrum in
  ## .createMsnbaseSpectrum2. validObject(msnexp) is really slow.
  ## spectra data should be correct and we don't interested in header
  ## information nor version information of the spectra
  ## see https://github.com/lgatto/MSnbase/issue/183
  #if (validObject(msnexp))
  #  return(msnexp)
  msnexp
}

#' create MS2 spectrum from spectrum.xml data
#' @param leID precursor.leID
#' @param ms1 ms1 matrix generated by .readSynapterSpectrumXml
#' @param ms2 ms2 matrix generated by .readSynapterSpectrumXml
#' @param leRows env generated by .readSynapterSpectrumXml
#' @param heRows env generated by .readSynapterSpectrumXml
#' @param removePrecursor remove precursor ion from fragments
#' @param tolerance tolerance to look for the precursor ion
#' @return Spectrum2 object
#' @noRd
.createMs2SpectrumFromSpectrumXml <- function(leID, ms1, ms2, leRows, heRows,
                                              removePrecursor=TRUE,
                                              tolerance=25e-6) {
  key <- as.character(leID)

  if (exists(key, envir=leRows)) {
    lrow <- get(key, envir=leRows)
    hrows <- get(key, envir=heRows)

    .createMsnbaseSpectrum2(leID=leID,
                            precursor.mhp=ms1[lrow, "Mass"],
                            precursor.inten=ms1[lrow, "Intensity"],
                            precursor.z=ms1[lrow, "Z"],
                            precursor.retT=ms1[lrow, "RT"],
                            mass=ms2[hrows, "Mass"],
                            intensity=ms2[hrows, "Intensity"],
                            removePrecursor=removePrecursor,
                            tolerance=tolerance)
  } else {
    .createEmptyMsnbaseSpectrum2(leID=leID)
  }
}

#' create MS2 spectrum from final_fragment.csv data
#' @param leID precursor.leID
#' @param fragments data.frame generated by readFragments
#' @param assignments env
#' @param removePrecursor remove precursor ion from fragments
#' @param tolerance tolerance to look for the precursor ion
#' @return Spectrum2 object
#' @noRd
.createMs2SpectrumFromFragments <- function(leID, fragments, assignments,
                                            removePrecursor=TRUE,
                                            tolerance=25e-6) {
  key <- as.character(leID)

  if (exists(key, envir=assignments)) {
    i <- get(key, envir=assignments)
    j <- i[1]

    .createMsnbaseSpectrum2(leID,
                            precursor.mhp=fragments$precursor.mhp[j],
                            precursor.inten=fragments$precursor.inten[j],
                            precursor.z=fragments$precursor.z[j],
                            precursor.retT=fragments$precursor.retT[j],
                            mass=fragments$product.mhp[i],
                            intensity=fragments$product.inten[i],
                            removePrecursor=removePrecursor,
                            tolerance=25e-6)
  } else {
    .createEmptyMsnbaseSpectrum2(leID=leID)
  }
}

#' create an instance of an MSnbase::Spectrum2 object
#' @param leID precursor.leID
#' @param precursor.mhp precursor mass
#' @param precursor.inten precursor intensity
#' @param precursor.z precursor charge
#' @param precursor.retT precursor retention time
#' @param mass mass/mz
#' @param intensity intensity
#' @param removePrecursor remove precursor ion from fragments
#' @param tolerance tolerance to look for the precursor ion
#' @return Spectrum2 object
#' @noRd
.createMsnbaseSpectrum2 <- function(leID, precursor.mhp, precursor.inten,
                                    precursor.z, precursor.retT,
                                    mass, intensity,
                                    removePrecursor=TRUE, tolerance=25e-6) {
  ## just to be sure, you can't trust any information in the csv/xml files
  o <- order(mass)

  if (removePrecursor) {
    m <- !as.logical(MSnbase:::relaxedMatch(x=mass[o], table=precursor.mhp,
                                            nomatch=0,
                                            relative=TRUE,
                                            tolerance=tolerance))
    o <- o[m]

    if (!length(o)) {
      return(.createEmptyMsnbaseSpectrum2(leID=leID))
    }
  }

  ## use MSnbase:::Spectrum2_mz_sorted to improve speed in class creation;
  ## see https://github.com/lgatto/MSnbase/issues/171
  #new("Spectrum2",
  #    precScanNum=as.integer(leID),
  #    precursorMz=precursor.mhp,
  #    precursorIntensity=precursor.inten,
  #    precursorCharge=as.integer(precursor.z),
  #    rt=precursor.retT,
  #    centroided=TRUE,
  #    # tic=precursor.inten,
  #    peaksCount=length(o),
  #    mz=mass[o], intensity=intensity[o],
  #    fromFile=1L)
  s <- MSnbase:::Spectrum2_mz_sorted(precScanNum=as.integer(leID),
                                     precursorMz=precursor.mhp,
                                     precursorIntensity=precursor.inten,
                                     precursorCharge=as.integer(precursor.z),
                                     rt=precursor.retT,
                                     centroided=TRUE,
                                     peaksCount=length(o),
                                     mz=mass[o], intensity=intensity[o],
                                     fromFile=1L)
  if (validObject(s)) {
    s
  }
}

#' create an empty instance of an MSnbase::Spectrum2 object
#' @param leID precursor.leID
#' @param key character (used in enviroments to access the correct spetrum)
#' @noRd
.createEmptyMsnbaseSpectrum2 <- function(leID, key="spectrum:-1") {
  if (missing(leID)) {
    leID <- as.integer(tail(strsplit(key, ":")[[1]], 1))
  }

  ## use MSnbase:::Spectrum2_mz_sorted to improve speed in class creation;
  ## see https://github.com/lgatto/MSnbase/issues/171
  #new("Spectrum2", precScanNum=as.integer(leID), fromFile=1L)
  MSnbase:::Spectrum2_mz_sorted(precScanNum=as.integer(leID), fromFile=1L)
}

#' @param msexp MSnExp object to filter
#' @param minIntensity minimal intensity
#' @param maxNumber maximal number of fragments to preserve
#' @param verbose verbose output?
#' @return modified MSnExp object
#' @noRd
.filterIntensity <- function(msexp, minIntensity=NULL, maxNumber=NULL,
                             verbose=interactive()) {
  if(is.null(minIntensity) && is.null(maxNumber)) {
    stop("At least one of arguments ", sQuote("minIntensity"), " and ",
         sQuote("maxNumber"), " must be given!")
  }

  minIntensity2 <- 0

  if (!is.null(maxNumber)) {
    cs <- .cumsumIntensities(msexp)
    minIntensity2 <- as.double(names(cs)[which(cs > (tail(cs, 1)-maxNumber))[1]])
  }

  minIntensity <- max(minIntensity, minIntensity2)

  if (verbose) {
    message("Set peaks with an intensity < ", minIntensity, " to zero")
    pcBefore <- .sumAllPeakCounts(msexp)
  }

  msexp <- removePeaks(msexp, t=minIntensity, verbose=verbose)

  ## remove corresponding fragment information if available
  stopifnot(fData(msexp)$leID == precScanNum(msexp))
  if ("fragment.str" %in% colnames(fData(msexp))) {
    fragments <- MSnbase:::utils.ssv2list(as.character(fData(msexp)$fragment.str))
    fData(msexp)$fragment.str <- MSnbase:::utils.list2ssv(mapply(function(f, i) {
      if (length(f) && length(i)) {
        as.character(f[i > 0])
      } else {
        character()
      }
    }, f=fragments, i=intensity(msexp), SIMPLIFY=FALSE, USE.NAMES=FALSE))
  }

  if (verbose) {
    message("Remove peaks with zero intensity")
    pcBefore <- .sumAllPeakCounts(msexp)
  }

  msexp <- clean(msexp, all=TRUE, verbose=verbose)

  if (verbose) {
    pcAfter <- .sumAllPeakCounts(msexp)
    message(pcBefore - pcAfter, " peaks removed; new total number of peaks: ",
            pcAfter)
  }
  msexp
}

#' @param msexp MSnExp object to filter
#' @noRd
.plotIntensityVsNumber <- function(msexp, what) {
  cs <- .cumsumIntensities(msexp)
  plot(as.double(names(cs)), cs, log="x", type="l",
       main=paste0("Cumulative Number of Fragments (", what, ")"),
       xlab="Intensity", ylab="# of Fragments")
  grid()
}

.cumsumIntensities <- function(msexp) {
  cumsum(table(unlist(intensity(msexp))))
}