R/matchSpectra.R

In michaelwitting/MetaboAnnotation: Utilities for Annotation of Metabolomics Data

#' @title Matching MS Spectra against a reference
#'
#' @aliases CompareSpectraParam-class MatchForwardReverseParam-class
#'
#' @name CompareSpectraParam
#'
#' @description
#'
#' `matchSpectra` compares experimental (*query*) MS2 spectra against
#' reference (*target*) MS2 spectra and reports matches with a similarity that
#' passing a specified threshold. The function performs the similarity
#' calculation between each query spectrum against each target spectrum.
#' Parameters `query` and `target` can be used to define the query and target
#' spectra, respectively, while parameter `param` allows to define and configure
#' the similarity calculation and matching condition. Parameter `query` takes
#' a [Spectra] object while `target` can be either a [Spectra] object, a
#' [CompDb] (reference library) object defined in the `CompoundDb` package or
#' a [CompAnnotationSource] (e.g. a [CompDbSource()])
#' with the reference or connection information to a supported annotation
#' resource).
#'
#' Some notes on performance and information on parallel processing are
#' provided in the vignette.
#'
#' Currently supported parameter objects defining the matching are:
#'
#' - `CompareSpectraParam`: the *generic* parameter object allowing to set all
#'   settings for the [compareSpectra()] call that is used to perform the
#'   similarity calculation. This includes `MAPFUN` and `FUN` defining the
#'   peak-mapping and similarity calculation functions and `ppm` and `tolerance`
#'   to define an acceptable difference between m/z values of the compared
#'   peaks. Additional parameters to the `compareSpectra` call
#'   can be passed along with `...`. See the help of [Spectra()] for more
#'   information on these parameters. Parameters `requirePrecursor` (default
#'   `TRUE`) and `requirePrecursorPeak` (default `FALSE`) allow to pre-filter
#'   the target spectra prior to the actual similarity calculation for each
#'   individual query spectrum. Target spectra can also be pre-filtered based on
#'   retention time if parameter `toleranceRt` is set to a value different than
#'   the default `toleranceRt = Inf`. Only target spectra with a retention time
#'   within the query's retention time +/- (`toleranceRt` + `percentRt`% of the
#'   query's retention time) are considered. Note that while for `ppm` and
#'   `tolerance` only a single value is accepted, `toleranceRt` and `percentRt`
#'   can be also of length equal to the number of query spectra hence allowing
#'   to define different rt boundaries for each query spectrum.
#'   While these pre-filters can considerably improve performance, it should be
#'   noted that no matches will be found between query and target spectra with
#'   missing values in the considered variable (precursor m/z or retention
#'   time). For target spectra without retention times (such as for `Spectra`
#'   from a public reference database such as MassBank) the default
#'   `toleranceRt = Inf` should thus be used.
#'   Finally, parameter `THRESHFUN` allows to define a function to be applied to
#'   the similarity scores to define which matches to report. See below for more
#'   details.
#'
#' - `MatchForwardReverseParam`: performs spectra matching as with
#'   `CompareSpectraParam` but reports, similar to MS-DIAL, also the *reverse*
#'   similarity score and the *presence ratio*. In detail, the matching of query
#'   spectra to target spectra is performed by considering all peaks from the
#'   query and all peaks from the target (reference) spectrum (i.e. *forward*
#'   matching using an *outer join*-based peak matching strategy). For matching
#'   spectra also the *reverse* similarity is calculated considering only peaks
#'   present in the target (reference) spectrum (i.e. using a *right join*-based
#'   peak matching). This is reported as spectra variable `"reverse_score"`.
#'   In addition, the ratio between the number of matched peaks and the total
#'   number of peaks in the target (reference) spectra is reported as the
#'   *presence ratio* (spectra variable `"presence_ratio"`) and the total
#'   number of matched peaks as `"matched_peaks_count"`. See examples below
#'   for details. Parameter `THRESHFUN_REVERSE` allows to define an additional
#'   *threshold function* to filter matches. If `THRESHFUN_REVERSE` is defined
#'   only matches with a spectra similarity fulfilling both `THRESHFUN` **and**
#'   `THRESHFUN_REVERSE` are returned. With the default
#'   `THRESHFUN_REVERSE = NULL` all matches passing `THRESHFUN` are reported.
#'
#' @param addOriginalQueryIndex for `matchSpectra()`: `logical(1)` whether an
#'   additional spectra variable `".original_query_index"` should be added to
#'   the `query` `Spectra` object providing the index of the spectrum in this
#'   originally provided object. This spectra variable can be useful to link
#'   back to the original `Spectra` object if the `MatchedSpectra` object gets
#'   subsetted/processed.
#'
#' @param BPPARAM for `matchSpectra`: parallel processing setup (see the
#'   `BiocParallel` package for more information). Parallel processing is
#'   disabled by default (with the default setting `BPPARAM = SerialParam()`).
#'
#' @param FUN `function` used to calculate similarity between spectra. Defaults
#'   for `CompareSpectraParam` to [MsCoreUtils::ndotproduct()]. See
#'   [MsCoreUtils::ndotproduct()] for details.
#'
#' @param MAPFUN `function` used to map peaks between the compared spectra.
#'    Defaults for `CompareSpectraParam` to [joinPeaks()]. See
#'    [compareSpectra()] for details.
#'
#' @param param for `matchSpectra`: parameter object (such as
#'   `CompareSpectraParam`) defining the settings for the matching.
#'
#' @param percentRt `numeric` of length 1 or equal to the number of query
#'   spectra defining the maximal accepted relative difference in retention
#'   time between query and target spectra expressed in percentage of the query
#'   rt. For `percentRt = 10`, similarities are defined between the query
#'   spectrum and all target spectra with a retention time within query rt
#'   +/- 10% of the query. By default (with `toleranceRt = Inf`) the retention
#'   time-based filter is not considered. Thus, to consider the `percentRt`
#'   parameter, `toleranceRt` should be set to a value different than that.
#'   See help of `CompareSpectraParam` above for more information.
#'
#' @param ppm `numeric(1)` for a relative, m/z-dependent, maximal accepted
#'   difference between m/z values. This will be used in `compareSpectra` as
#'   well as for eventual precursor m/z matching.
#'
#' @param query for `matchSpectra`: [Spectra] object with the query spectra.
#'
#' @param requirePrecursor `logical(1)` whether only target spectra are
#'   considered in the similarity calculation with a precursor m/z that matches
#'   the precursor m/z of the query spectrum (considering also `ppm` and
#'   `tolerance`). With `requirePrecursor = TRUE` (the default) the function
#'   will complete much faster, but will not find any hits for target (or query
#'   spectra) with missing precursor m/z. It is suggested to check first the
#'   availability of the precursor m/z in `target` and `query`.
#'
#' @param requirePrecursorPeak `logical(1)` whether only target spectra will be
#'   considered in the spectra similarity calculation that have a peak with an
#'   m/z matching the precursor m/z of the query spectrum. Defaults to
#'   `requirePrecursorPeak = FALSE`. It is suggested to check first the
#'   availability of the precursor m/z in `query`, as no match will be reported
#'   for query spectra with missing precursor m/z.
#'
#' @param rtColname `character(2)` with the name of the spectra variable
#'     containing the retention time information for compounds to be used in
#'     retention time matching (only used if `toleranceRt` is not `Inf`).
#'     It can also be `character(1)` if the two names are the same.
#'     Defaults to `rtColname = c("rtime", "rtime")`.
#'
#' @param target for `matchSpectra`: [Spectra], [CompDb] or object extending
#'   [CompAnnotationSource] (such as [CompDbSource]) with
#'   the target (reference) spectra to compare `query` against.
#'
#' @param tolerance `numeric(1)` for an absolute maximal accepted difference
#'   between m/z values. This will be used in `compareSpectra` as well as for
#'   eventual precursor m/z matching.
#'
#' @param toleranceRt `numeric` of length 1 or equal to the number of query
#'   spectra defining the maximal accepted (absolute) difference in retention
#'   time between query and target spectra. By default
#'   (with `toleranceRt = Inf`) the retention time-based filter is not
#'   considered. See help of `CompareSpectraParam` above for more
#'   information.
#'
#' @param THRESHFUN `function` applied to the similarity score to define which
#'   target spectra are considered *matching*. Defaults to
#'   `THRESHFUN = function(x) which(x >= 0.7)` hence selects
#'   all target spectra matching a query spectrum with a similarity higher or
#'   equal than `0.7`. Any function that takes a numeric vector with similarity
#'   scores from the comparison of a query spectrum with all target spectra (as
#'   returned by [compareSpectra()]) as input and returns a
#'   `logical` vector (same dimensions as the similarity scores) or an integer
#'   with the matches is supported.
#'
#' @param THRESHFUN_REVERSE for `MatchForwardReverseParam`: optional additional
#'   *thresholding function* to filter the results on the reverse score. If
#'   specified the same format than `THRESHFUN` is expected.
#'
#' @param ... for `CompareSpectraParam`: additional parameters passed along
#'   to the [compareSpectra()] call.
#'
#' @author Johannes Rainer, Michael Witting
#'
#' @importClassesFrom ProtGenerics Param
#'
#' @importClassesFrom Spectra Spectra
#'
#' @importFrom MsCoreUtils ndotproduct
#'
#' @importFrom Spectra joinPeaks
#'
#' @rdname CompareSpectraParam
#'
#' @return `matchSpectra` returns a [MatchedSpectra()] object with the matching
#'   results. If `target` is a `CompAnnotationSource` only matching target
#'   spectra will be reported.
#'
#'   Constructor functions return an instance of the class.
#'
#' @exportClass CompareSpectraParam
#'
#' @examples
#'
#' library(Spectra)
#' library(msdata)
#' fl <- system.file("TripleTOF-SWATH", "PestMix1_DDA.mzML", package = "msdata")
#' pest_ms2 <- filterMsLevel(Spectra(fl), 2L)
#'
#' ## subset to selected spectra.
#' pest_ms2 <- pest_ms2[c(808, 809, 945:955)]
#'
#' ## Load a small example MassBank data set
#' load(system.file("extdata", "minimb.RData", package = "MetaboAnnotation"))
#'
#' ## Match spectra with the default similarity score (normalized dot product)
#' csp <- CompareSpectraParam(requirePrecursor = TRUE, ppm = 10)
#' mtches <- matchSpectra(pest_ms2, minimb, csp)
#'
#' mtches
#'
#' ## Are there any matching spectra for the first query spectrum?
#' mtches[1]
#' ## No
#'
#' ## And for the second query spectrum?
#' mtches[2]
#' ## The second query spectrum matches 4 target spectra. The scores for these
#' ## matches are:
#' mtches[2]$score
#'
#' ## To access the score for the full data set
#' mtches$score
#'
#' ## Below we use a THRESHFUN that returns for each query spectrum the (first)
#' ## best matching target spectrum.
#' csp <- CompareSpectraParam(requirePrecursor = FALSE, ppm = 10,
#'     THRESHFUN = function(x) which.max(x))
#' mtches <- matchSpectra(pest_ms2, minimb, csp)
#' mtches
#'
#' ## Each of the query spectra is matched to one target spectrum
#' length(mtches)
#' matches(mtches)
#'
#' ## Match spectra considering also measured retention times. This requires
#' ## that both query and target spectra have non-missing retention times.
#' rtime(pest_ms2)
#' rtime(minimb)
#'
#' ## Target spectra don't have retention times. Below we artificially set
#' ## retention times to show how an additional retention time filter would
#' ## work.
#' rtime(minimb) <- rep(361, length(minimb))
#'
#' ## Matching spectra requiring a matching precursor m/z and the difference
#' ## of retention times between query and target spectra to be <= 2 seconds.
#' csp <- CompareSpectraParam(requirePrecursor = TRUE, ppm = 10,
#'     toleranceRt = 2)
#' mtches <- matchSpectra(pest_ms2, minimb, csp)
#' mtches
#' matches(mtches)
#'
#' ## Note that parameter `rtColname` can be used to define different spectra
#' ## variables with retention time information (such as retention indices etc).
#'
#' ## A `CompDb` compound annotation database could also be used with
#' ## parameter `target`. Below we load the test `CompDb` database from the
#' ## `CompoundDb` Bioconductor package.
#' library(CompoundDb)
#' fl <- system.file("sql", "CompDb.MassBank.sql", package = "CompoundDb")
#' cdb <- CompDb(fl)
#' res <- matchSpectra(pest_ms2, cdb, CompareSpectraParam())
#'
#' ## We do however not find any matches since the used compound annotation
#' ## database contains only a very small subset of the MassBank.
#' res
#'
#' ## As `target` we have now however the MS2 spectra data from the compound
#' ## annotation database
#' target(res)
#'
#' ## See the package vignette for details, descriptions and more examples,
#' ## also on how to retrieve e.g. MassBank reference databases from
#' ## Bioconductor's AnnotationHub.
NULL

setClass("CompareSpectraParam",
         slots = c(
             MAPFUN = "function",
             tolerance = "numeric",
             ppm = "numeric",
             FUN = "function",
             dots = "list",
             requirePrecursor = "logical",
             requirePrecursorPeak = "logical",
             THRESHFUN = "function",
             toleranceRt = "numeric",
             percentRt = "numeric"
             ),
         contains = "Param",
         prototype = prototype(
             MAPFUN = joinPeaks,
             tolerance = 0,
             ppm = 5,
             FUN = MsCoreUtils::ndotproduct,
             dots = list(),
             requirePrecursor = TRUE,
             requirePrecursorPeak = FALSE,
             THRESHFUN = function(x) which(x >= 0.7),
             toleranceRt = Inf,
             percentRt = 0
         ),
         validity = function(object) {
             msg <- NULL
             if (length(object@tolerance) > 1 || object@tolerance < 0)
                 msg <- c("'tolerance' has to be a positive number of length 1")
             if (length(object@ppm) > 1 || object@ppm < 0)
                 msg <- c("'ppm' has to be positive number of length 1")
             msg <- c(msg, .valid_threshfun(object@THRESHFUN))
             if (any(object@toleranceRt < 0))
                 msg <- c("'toleranceRt' has to be positive")
             if (any(object@percentRt < 0))
                 msg <- c("'percentRt' has to be positive")
             msg
         })

setClass("MatchForwardReverseParam",
         slots = c(THRESHFUN_REVERSE = "ANY"),
         contains = "CompareSpectraParam",
         prototype = prototype(
             THRESHFUN_REVERSE = NULL
         ),
         validity = function(object) {
             .valid_threshfun(object@THRESHFUN_REVERSE, "THRESHFUN_REVERSE")
         })

#' @rdname CompareSpectraParam
#'
#' @importMethodsFrom Spectra compareSpectra spectraNames<- precursorMz
#'
#' @importFrom MsCoreUtils ppm
#'
#' @importFrom BiocParallel bplapply
#'
#' @importFrom methods new
#'
#' @export
CompareSpectraParam <- function(MAPFUN = joinPeaks, tolerance = 0, ppm = 5,
                                FUN = MsCoreUtils::ndotproduct,
                                requirePrecursor = TRUE,
                                requirePrecursorPeak = FALSE,
                                THRESHFUN = function(x) which(x >= 0.7),
                                toleranceRt = Inf, percentRt = 0, ...) {
    new("CompareSpectraParam", MAPFUN = force(MAPFUN), tolerance = tolerance,
        ppm = ppm, FUN = force(FUN), requirePrecursor = requirePrecursor[1L],
        requirePrecursorPeak = requirePrecursorPeak[1L],
        THRESHFUN = force(THRESHFUN), toleranceRt = toleranceRt,
        percentRt = percentRt, dots = list(...))
}

#' @rdname CompareSpectraParam
#'
#' @export
MatchForwardReverseParam <- function(MAPFUN = joinPeaks, tolerance = 0, ppm = 5,
                                     FUN = MsCoreUtils::ndotproduct,
                                     requirePrecursor = TRUE,
                                     requirePrecursorPeak = FALSE,
                                     THRESHFUN = function(x) which(x >= 0.7),
                                     THRESHFUN_REVERSE = NULL,
                                     toleranceRt = Inf, percentRt = 0, ...) {
    dots <- list(...)
    if (any(names(dots) == "type")) {
        warning("Specifying a join type with parameter 'type' is not ",
                "supported. Will ignore parameter 'type'")
        dots$type <- NULL
    }
    new("MatchForwardReverseParam", MAPFUN = MAPFUN, tolerance = tolerance,
        ppm = ppm, FUN = FUN, requirePrecursor = requirePrecursor[1L],
        requirePrecursorPeak = requirePrecursorPeak[1L],
        THRESHFUN = THRESHFUN, THRESHFUN_REVERSE = THRESHFUN_REVERSE,
        toleranceRt = toleranceRt, percentRt = percentRt, dots = dots)
}

.valid_threshfun <- function(x, variable = "THRESHFUN") {
    msg <- NULL
    if (length(x)) {
        if (!is.function(x))
            return(paste0(variable, " needs to be a function."))
        res <- x(1:10)
        if (!is.logical(res) & !is.integer(res))
            return(paste0(variable,
                          " needs to return integer or logical vector."))
        if (is.logical(res) && length(res) != 10)
            return(paste0(variable, " needs to return a logical vector with",
                          " the same length than target spectra."))
        if (is.integer(res) && any(res < 1 | res > 10))
            return(paste0(variable, " needs to return an integer vector with ",
                          "values between 1 and the length of target spectra."))
    }
    msg
}

.compare_spectra_parms_list <- function(x) {
    c(list(MAPFUN = x@MAPFUN, tolerance = x@tolerance, ppm = x@ppm,
           FUN = x@FUN), x@dots)
}

#' @importMethodsFrom Spectra spectraNames containsMz filterPrecursorMzRange
#'
#' @importMethodsFrom Spectra backendBpparam
#'
#' @rdname CompareSpectraParam
#'
#' @export
setMethod(
    "matchSpectra",
    signature(query = "Spectra", target = "Spectra",
              param = "CompareSpectraParam"),
    function(query, target, param, rtColname = c("rtime", "rtime"),
             BPPARAM = BiocParallel::SerialParam(),
             addOriginalQueryIndex = TRUE) {
        BPPARAM <- .check_bpparam(query, target, BPPARAM)
        if (addOriginalQueryIndex) {
            if (any(spectraVariables(query) == ".original_query_index"))
                warning("Overwriting already present spectra variable ",
                        "\".original_query_index\"")
            query$.original_query_index <- seq_along(query)
        }
        if (length(query) == 1 || param@requirePrecursor ||
            param@requirePrecursorPeak || any(is.finite(param@toleranceRt)) ||
            any(param@percentRt != 0))
            .match_spectra(query, target, param, rtColname = rtColname, BPPARAM)
        else .match_spectra_without_precursor(query, target, param)
    })

#' Returns SerialParam if any of the backends does not support parallel
#' processing.
#'
#' @noRd
.check_bpparam <- function(query, target, BPPARAM) {
    BPPARAM <- backendBpparam(query, BPPARAM)
    if (!is(BPPARAM, "SerialParam"))
        BPPARAM <- backendBpparam(target, BPPARAM)
    BPPARAM
}

#' @importClassesFrom CompoundDb CompDb
#'
#' @rdname CompareSpectraParam
#'
#' @export
setMethod(
    "matchSpectra", signature(query = "Spectra", target = "CompDb",
                              param = "Param"),
    function(query, target, param, rtColname = c("rtime", "rtime"),
             BPPARAM = BiocParallel::SerialParam(),
             addOriginalQueryIndex = TRUE) {
        matchSpectra(query, Spectra(target), param = param,
                     rtColname = rtColname, BPPARAM = BPPARAM,
                     addOriginalQueryIndex = addOriginalQueryIndex)
    })

.match_spectra <- function(query, target, param,
                           rtColname = c("rtime", "rtime"), BPPARAM) {
    if (length(rtColname) != 2)
        rtColname <- rep(rtColname[1L], 2)
    parms <- .compare_spectra_parms_list(param)
    queryl <- length(query)
    toleranceRt <- param@toleranceRt
    percentRt <- param@percentRt
    if (length(toleranceRt) == 1L)
        toleranceRt <- rep(toleranceRt, queryl)
    if (length(percentRt) == 1L)
        percentRt <- rep(percentRt, queryl)
    if (length(percentRt) != queryl || length(toleranceRt) != queryl)
        stop("Length of 'toleranceRt' and 'percentRt' has to be either 1 or ",
             "equal to the number of query spectra")
    if (is.null(spectraNames(target)))
        spectraNames(target) <- seq_along(target)
    snames <- spectraNames(target)
    maps <- bplapply(seq_along(query), .get_matches_spectra,
                     query = query,
                     target = target,
                     parlist = parms,
                     THRESHFUN = param@THRESHFUN,
                     precMz = param@requirePrecursor,
                     precMzPeak = param@requirePrecursorPeak,
                     toleranceRt = toleranceRt,
                     percentRt = percentRt,
                     query_rt_col = rtColname[1L],
                     target_rt_col = rtColname[2L],
                     sn = snames, BPPARAM = BPPARAM)
    maps <- do.call(rbind.data.frame, maps)
    if (!nrow(maps))
        maps <- data.frame(query_idx = integer(),
                           target_idx = integer(),
                           score = numeric())
    .matched_spectra(query = query, target = target, matches = maps,
                     metadata = list(param = param), validate = FALSE)
}

#' This version does not use a loop but performs the comparison all in one. It
#' can thus only be performed if no precursor check/filter is used.
#'
#' @noRd
.match_spectra_without_precursor <- function(query, target, param) {
    parlist <- .compare_spectra_parms_list(param)
    cor <- do.call(compareSpectra, c(list(x = query, y = target), parlist))
    res <- lapply(seq_len(nrow(cor)), function(i) param@THRESHFUN(cor[i, ]))
    if (is.logical(res[[1L]]))
        res <- lapply(res, which)
    tidx <- unlist(res, use.names = FALSE)
    if (length(tidx)) {
        qidx <- rep(seq_along(res), lengths(res))
        maps <- data.frame(query_idx = qidx, target_idx = tidx,
                           score = cor[cbind(qidx, tidx)])
    }
    else maps <- data.frame(query_idx = integer(), target_idx = integer(),
                            score = numeric())
    .matched_spectra(query = query, target = target, matches = maps,
                     metadata = list(param = param), validate = FALSE)
}

#' @importMethodsFrom Spectra filterRt rtime spectraData
#'
#' @importFrom MsCoreUtils between
#'
#' @return `data.frame` with matches or `NULL` (if not matches passed the
#'     filter).
#'
#' @noRd
.get_matches_spectra <- function(i, query, target, parlist, THRESHFUN, precMz,
                                 precMzPeak, toleranceRt = Inf,
                                 percentRt = 0, sn, query_rt_col = "rtime",
                                 target_rt_col = "rtime") {
    qi <- query[i]
    if (is.finite(toleranceRt[i])) {
        rt_qi <- spectraData(qi, query_rt_col)[, 1L]
        if (toleranceRt[i] == 0 && percentRt[i] == 0)
            trt <- 0.0000001
        else
            trt <- toleranceRt[i] + rt_qi * percentRt[i] / 100
        if (target_rt_col != "rtime") {
            rt_t <- spectraData(target, target_rt_col)[, 1L]
            target <- target[between(rt_t, rt_qi + c(-trt, trt))]
        } else target <- filterRt(target, rt = rt_qi + c(-trt, trt))
    }
    if (precMz) {
        pmz <- precursorMz(qi)
        pmz <- pmz + c(-1, 1) * (ppm(pmz, parlist$ppm) + parlist$tolerance)
        target <- filterPrecursorMzRange(target, mz = pmz)
    }
    if (precMzPeak)
        target <- target[containsMz(target, mz = precursorMz(qi),
                                ppm = parlist$ppm,
                                tolerance = parlist$tolerance)]
    if (!length(target))
        return(NULL)
    cor <- base::do.call(compareSpectra,
                         c(list(x = qi, y = target, SIMPLIFY = FALSE), parlist))
    keep <- THRESHFUN(as.vector(cor))
    if (is.logical(keep))
        keep <- which(keep)
    kl <- length(keep)
    if (kl)
        data.frame(query_idx = rep(i, kl),
                   target_idx = base::match(spectraNames(target)[keep], sn),
                   score = cor[keep])
    else NULL
}

#' @rdname CompareSpectraParam
#'
#' @importFrom methods as
#'
#' @importMethodsFrom Spectra peaksData
#'
#' @export
setMethod(
    "matchSpectra",
    signature(query = "Spectra", target = "Spectra",
              param = "MatchForwardReverseParam"),
    function(query, target, param, rtColname = c("rtime", "rtime"),
             BPPARAM = BiocParallel::SerialParam(),
             addOriginalQueryIndex = TRUE) {
        res <- matchSpectra(query, target, as(param, "CompareSpectraParam"),
                            rtColname = rtColname, BPPARAM = BPPARAM,
                            addOriginalQueryIndex = addOriginalQueryIndex)
        ## Loop over the matches and assign additional stuff...
        nm <- nrow(res@matches)
        res@matches$reverse_score <- rep(NA_real_, nm)
        res@matches$presence_ratio <- rep(NA_real_, nm)
        res@matches$matched_peaks_count <- rep(NA_real_, nm)
        parms_rv <- .compare_spectra_parms_list(param)
        parms_rv$type <- "right"
        for (i in seq_len(nm)) {
            spl <- c(
                list(x = peaksData(query[res@matches$query_idx[i]])[[1L]],
                     y = peaksData(target[res@matches$target_idx[i]])[[1L]]),
                parms_rv)
            map <- do.call(param@MAPFUN, spl)
            spl$x <- map$x
            spl$y <- map$y
            cor <- do.call(param@FUN, spl)
            res@matches$reverse_score[i] <- cor
            nmatched <- sum(!is.na(map$x[, 1L]))
            res@matches$presence_ratio[i] <- nmatched /
                nrow(map$y)
            res@matches$matched_peaks_count[i] <- nmatched
        }
        if (length(param@THRESHFUN_REVERSE))
            res@matches <- res@matches[param@THRESHFUN_REVERSE(
                                                 res@matches$reverse_score), ,
                                       drop = FALSE]
        res@metadata$param <- param
        res
    })