Spectra: Spectra Infrastructure for Mass Spectrometry Data

Documented in combinePeaks

#' @title Function dealing with peak lists
#'
#' @description
#'
#' These functions are supposed to take a `list` of peak matrices (i.e. two
#' column matrices with columns `"mz"` and `"intensity"` as input and return
#' a single `matrix` combining the individual input peak matrices.
#'
#' @noRd
NULL

#' @title Combine peaks with similar m/z across spectra
#'
#' @description
#'
#' `combinePeaks` aggregates provided peak matrices into a single peak matrix.
#' Peaks are grouped by their m/z values with the `group()` function from the
#' `MsCoreUtils` package. In brief, all peaks in all provided spectra are first
#' ordered by their m/z and consecutively grouped into one group if the
#' (pairwise) difference between them is smaller than specified with parameter
#' `tolerance` and `ppm` (see [group()] for grouping details and examples).
#'
#' The m/z and intensity values for the resulting peak matrix are calculated
#' using the `mzFun` and `intensityFun` on the grouped m/z and intensity values.
#'
#' The function supports also different strategies for peak combinations which
#' can be specified with the `peaks` parameter:
#'
#' - `peaks = "union"` (default): report all peaks from all input spectra.
#'
#' - `peaks = "intersect"`: keep only peaks in the resulting peak matrix that
#'   are present in `>= minProp` proportion of input spectra. This would
#'   generate a *consensus* or *representative* spectra from a set of e.g.
#'   fragment spectra measured from the same precursor ion.
#'
#' As a special case it is possible to report only peaks in the resulting
#' matrix from peak groups that contain a peak from one of the input spectra,
#' which can be specified with parameter `main`. Thus, if e.g. `main = 2` is
#' specified, only (grouped) peaks that have a peak in the second input matrix
#' are returned.
#'
#' Setting `timeDomain` to `TRUE` causes grouping to be performed on the square
#' root of the m/z values (assuming a TOF instrument was used to create the
#' data).
#'
#' @details
#'
#' For general merging of spectra, the `tolerance` and/or `ppm` should be
#' manually specified based on the precision of the MS instrument. Peaks
#' from spectra with a difference in their m/z being smaller than `tolerance`
#' or smaller than `ppm` of their m/z are grouped into the same final peak.
#'
#' Some details for the combination of consecutive spectra of an LC-MS run:
#'
#' The m/z values of the same ion in consecutive scans (spectra) of a LC-MS run
#' will not be identical. Assuming that this random variation is much smaller
#' than the resolution of the MS instrument (i.e. the difference between
#' m/z values within each single spectrum), m/z value groups are defined
#' across the spectra and those containing m/z values of the `main` spectrum
#' are retained.
#' Intensities and m/z values falling within each of these m/z groups are
#' aggregated using the `intensityFun` and `mzFun`, respectively. It is
#' highly likely that all QTOF profile data is collected with a timing circuit
#' that collects data points with regular intervals of time that are then later
#' converted into m/z values based on the relationship `t = k * sqrt(m/z)`. The
#' m/z scale is thus non-linear and the m/z scattering (which is in fact caused
#' by small variations in the time circuit) will thus be different in the lower
#' and upper m/z scale. m/z-intensity pairs from consecutive scans to be
#' combined are therefore defined by default on the square root of the m/z
#' values. With `timeDomain = FALSE`, the actual m/z values will be used.
#'
#' @param x `list` of peak matrices.
#'
#' @param intensityFun `function` to be used to combine intensity values for
#'     matching peaks. By default the mean intensity value is returned.
#'
#' @param peaks `character(1)` specifying how peaks should be combined. Can be
#'     either `"peaks = "union"` (default) or `peaks = "intersect"`. See
#'     function description for details.
#'
#' @param minProp `numeric(1)` for `peaks = "intersect": the minimal required
#'     proportion of input spectra (peak matrices) a mass peak has to be
#'     present to be included in the consensus peak matrix.
#'
#' @param mzFun `function` to be used to combine m/z values for matching peaks.
#'     By default the mean m/z value is returned.
#'
#' @param weighted `logical(1)` defining whether m/z values for matching peaks
#'     should be calculated by an intensity-weighted average of the individuak
#'     m/z values. This overrides parameter `mzFun`.
#'
#' @param tolerance `numeric(1)` defining the (absolute) maximal accepted
#'     difference between mass peaks to group them into the same final peak.
#'
#' @param ppm `numeric(1)` defining the m/z-relative maximal accepted
#'     difference between mass peaks (expressed in parts-per-million) to group
#'     them into the same final peak.
#'
#' @param timeDomain `logical(1)` whether grouping of mass peaks is performed
#'     on the m/z values (`timeDomain = FALSE`) or on `sqrt(mz)`
#'     (`timeDomain = TRUE`).
#'
#' @param main optional `integer(1)` to force the resulting peak list to
#'     contain only peaks that are present in the specified input spectrum. See
#'     description for details.
#'
#' @param ... additional parameters to the `mzFun` and `intensityFun` functions.
#'
#' @family peak matrix combining functions
#'
#' @return
#'
#' Peaks `matrix` with m/z and intensity values representing the aggregated
#' values across the provided peak matrices.
#'
#' @author Johannes Rainer
#'
#' @importFrom MsCoreUtils group
#'
#' @importFrom stats weighted.mean
#'
#' @export
#'
#' @examples
#'
#' set.seed(123)
#' mzs <- seq(1, 20, 0.1)
#' ints1 <- abs(rnorm(length(mzs), 10))
#' ints1[11:20] <- c(15, 30, 90, 200, 500, 300, 100, 70, 40, 20) # add peak
#' ints2 <- abs(rnorm(length(mzs), 10))
#' ints2[11:20] <- c(15, 30, 60, 120, 300, 200, 90, 60, 30, 23)
#' ints3 <- abs(rnorm(length(mzs), 10))
#' ints3[11:20] <- c(13, 20, 50, 100, 200, 100, 80, 40, 30, 20)
#'
#' ## Create the peaks matrices
#' p1 <- cbind(mz = mzs + rnorm(length(mzs), sd = 0.01),
#'     intensity = ints1)
#' p2 <- cbind(mz = mzs + rnorm(length(mzs), sd = 0.01),
#'     intensity = ints2)
#' p3 <- cbind(mz = mzs + rnorm(length(mzs), sd = 0.009),
#'     intensity = ints3)
#'
#' ## Combine the spectra. With `tolerance = 0` and `ppm = 0` only peaks with
#' ## **identical** m/z are combined. The result will be a single spectrum
#' ## containing the *union* of mass peaks from the individual input spectra.
#' p <- combinePeaks(list(p1, p2, p3))
#'
#' ## Plot the spectra before and after combining
#' par(mfrow = c(2, 1), mar = c(4.3, 4, 1, 1))
#' plot(p1[, 1], p1[, 2], xlim = range(mzs[5:25]), type = "h", col = "red")
#' points(p2[, 1], p2[, 2], type = "h", col = "green")
#' points(p3[, 1], p3[, 2], type = "h", col = "blue")
#'
#' plot(p[, 1], p[, 2], xlim = range(mzs[5:25]), type = "h",
#'     col = "black")
#' ## The peaks were not merged, because their m/z differs too much.
#'
#' ## Combine spectra with `tolerance = 0.05`. This will merge all triplets.
#' p <- combinePeaks(list(p1, p2, p3), tolerance = 0.05)
#'
#' ## Plot the spectra before and after combining
#' par(mfrow = c(2, 1), mar = c(4.3, 4, 1, 1))
#' plot(p1[, 1], p1[, 2], xlim = range(mzs[5:25]), type = "h", col = "red")
#' points(p2[, 1], p2[, 2], type = "h", col = "green")
#' points(p3[, 1], p3[, 2], type = "h", col = "blue")
#'
#' plot(p[, 1], p[, 2], xlim = range(mzs[5:25]), type = "h",
#'     col = "black")
#'
#' ## With `intensityFun = max` the maximal intensity per peak is reported.
#' p <- combinePeaks(list(p1, p2, p3), tolerance = 0.05,
#'     intensityFun = max)
#'
#' ## Create *consensus*/representative spectrum from a set of spectra
#'
#' p1 <- cbind(mz = c(12, 45, 64, 70), intensity = c(10, 20, 30, 40))
#' p2 <- cbind(mz = c(17, 45.1, 63.9, 70.2), intensity = c(11, 21, 31, 41))
#' p3 <- cbind(mz = c(12.1, 44.9, 63), intensity = c(12, 22, 32))
#'
#' ## No mass peaks identical thus consensus peaks are empty
#' combinePeaks(list(p1, p2, p3), peaks = "intersect")
#'
#' ## Reducing the minProp to 0.2. The consensus spectrum will contain all
#' ## peaks
#' combinePeaks(list(p1, p2, p3), peaks = "intersect", minProp = 0.2)
#'
#' ## With a tolerance of 0.1 mass peaks can be matched across spectra
#' combinePeaks(list(p1, p2, p3), peaks = "intersect", tolerance = 0.1)
#'
#' ## Report the minimal m/z and intensity
#' combinePeaks(list(p1, p2, p3), peaks = "intersect", tolerance = 0.1,
#'     intensityFun = min, mzFun = min)
combinePeaks <- function(x, intensityFun = base::mean,
                         mzFun = base::mean, weighted = FALSE,
                         tolerance = 0, ppm = 0, timeDomain = FALSE,
                         peaks = c("union", "intersect"), main = integer(),
                         minProp = 0.5, ...) {
    peaks <- match.arg(peaks)
    lenx <- length(x)
    if (lenx == 1)
        return(x[[1]])
    mzs <- lapply(x, "[", , y = 1)
    mzs_lens <- lengths(mzs)
    mzs <- unlist(mzs, use.names = FALSE)
    mz_order <- order(mzs)
    mzs <- mzs[mz_order]
    if (timeDomain)
        mz_groups <- group(sqrt(mzs), tolerance = tolerance, ppm = ppm)
    else
        mz_groups <- group(mzs, tolerance = tolerance, ppm = ppm)
    ints <- unlist(lapply(x, "[", , y = 2), use.names = FALSE)[mz_order]
    if (length(main)) {
        if (main < 1 || main > lenx)
            stop("'main' has to be larger than 1 and smaller than ", lenx)
        ## Keep only mz groups present in the *main* spectrum
        is_in_main <- rep.int(seq_along(mzs_lens), mzs_lens)[mz_order] == main
        keep <- mz_groups %in% mz_groups[is_in_main]
        ## Keep only values for which a m/z in main is present.
        mz_groups <- mz_groups[keep]
        mzs <- mzs[keep]
        ints <- ints[keep]
    }
    mzs <- split(mzs, mz_groups)
    ints <- split(ints, mz_groups)
    if (peaks == "intersect") {
        keep <- lengths(mzs) >= (lenx * minProp)
        if (any(keep)) {
            mzs <- mzs[keep]
            ints <- ints[keep]
        } else
            return(cbind(mz = numeric(), intensity = numeric()))
    }
    if (weighted) {
        wm <- stats::weighted.mean
        mzs <- mapply(mzs, ints, FUN = function(mz, w)
                      wm(mz, w + 1, na.rm = TRUE, USE.NAMES = FALSE))
    } else
        mzs <- vapply1d(mzs, FUN  = mzFun)
    if (is.unsorted(mzs))
        stop("m/z values of combined spectrum are not ordered increasingly")
    cbind(mz = mzs, intensity = vapply1d(ints, FUN = intensityFun))
}

#' pairwise comparison of peak matrices. Parameters `xPrecursorMz` and
#' `yPrecursorMz` are passed to the `MAPFUN` and to `FUN`.
#'
#' @param x `list` of peak matrix
#'
#' @param y `list` of peak matrix
#'
#' @param xPrecursorMz `numeric` with the precursor m/z of each spectrum in `x`.
#'
#' @param yPrecursorMz `numeric` with the precursor m/z of each spectrum in `y`.
#'
#' @inheritParams .compare_spectra_chunk
#'
#' @return `matrix` with similarities.
#'
#' @author Sebastian Gibb, Johannes Rainer
#'
#' @noRd
.peaks_compare <- function(x, y, MAPFUN = joinPeaks, tolerance = 0,
                           ppm = 20, FUN = ndotproduct, xPrecursorMz,
                           yPrecursorMz, ...) {
    mat <- matrix(NA_real_, nrow = length(x), ncol = length(y),
                  dimnames = list(names(x), names(y)))
    for (i in seq_along(x)) {
        for (j in seq_along(y)) {
            peak_map <- MAPFUN(x[[i]], y[[j]],
                               tolerance = tolerance, ppm = ppm,
                               xPrecursorMz = xPrecursorMz[i],
                               yPrecursorMz = yPrecursorMz[j],
                               .check = FALSE, ...)
            mat[i, j] <- FUN(peak_map[[1L]], peak_map[[2L]],
                             xPrecursorMz = xPrecursorMz[i],
                             yPrecursorMz = yPrecursorMz[j], ...)
        }
    }
    mat
}