R/correctUnmix.R

In jtheorell/flowSpecs: Tools for processing of high-dimensional cytometry data

#' Correct defects in spectral unmixing by compensation
#'
#'
#' This function provides a way to reduce the defects in the spectral unmixing,
#' by creating a secondary correction matrix, which is symmetrical.
#' @param unmixFlowObj A flowframe or flowset post unmixing.
#' @param corrMat A correction matrix. If this is the first round, the
#' executionof this function needs to be preceeded by the generation of this
#' matrix, for example by using the \code{\link{corrMatCreate}} function.
#' @param transCoFacs If transformation should be performed, the transformation
#' cofactors can be added here. Three possible inputs: a vector with specific
#' cofactors for each variable, a set value that will be used for all variables,
#' and FALSE.
#' Note: It might be good to set this to FALSE in the final round, to optimize
#' the transoformations externally.
#' @return The unmixed flow object, now corrected with the values from the
#' corrMat.
#' @seealso \code{\link{specUnmix}}, \code{\link{arcTrans}},
#' \code{\link{corrMatCreate}}
#' @examples
#' # Load uncompensated data
#' data(fullPanel)
#'
#' # Load the spectral unmixing matrix generated with controls from the same
#' # experiment. These can be generated using the specMatCalc function.
#' data(specMat)
#'
#' # And now unmix
#' fullPanelUnmix <- specUnmix(fullPanel, specMat)
#'
#'
#' # Create an empty unmixinng matrix
#' corrMat <- corrMatCreate(specMat)
#'
#' # Now correct the data with this. In the first instance, this will of course
#' # not have any effect, more  than transformation, as the corrMat is empty.
#' fullPanelCorr <- correctUnmix(fullPanelUnmix, corrMat)
#'
#' # This now needs to be investigated, to identify any possible compensation
#' # defects. This is most easily done with the oneVsAllPlot executed in the
#' # following way:
#' \dontrun{
#' oneVsAllPlot(fullPanelCorr)
#' }
#' # One obvoius defect that shows when doing this is between CD56 and IgM:
#' oneVsAllPlot(fullPanelCorr, "BV650_CD56", saveResult = FALSE)
#'
#' # This is corrcted the following way:
#' corrMat["BV650_CD56", "AF647_IgM"] <- -0.03
#' fullPanelCorr <- correctUnmix(fullPanelUnmix, corrMat)
#' oneVsAllPlot(fullPanelCorr, "BV650_CD56", saveResult = FALSE)
#'
#' # This process is iterated until there are no remaining artifacts. Good help
#' # to do this is a set of fluorescence-minus-one controls. If that is not
#' # available, a rule of thumb is that if the signal in marker x is
#' # strongly negatively correlated to marker y, so that highly
#' # single-x-posisive values are below zero, then this is with all likelihood
#' # an artifact. The situation becomes more complicated with strong positive
#' # correlations, as they can occur in biology, so there one has to take more
#' # care and keep the marker biology in mind.
#'
#' @export correctUnmix
correctUnmix <- function(unmixFlowObj, corrMat, transCoFacs = 400) {
    corrUnmixFlowObj <- specUnmix(unmixFlowObj, corrMat)

    if (is.logical(transCoFacs) && transCoFacs == FALSE) {
        return(corrUnmixFlowObj)
    } else if (is.numeric(transCoFacs)) {
        if (length(transCoFacs) == 1) {
            transCoFacs <- rep(transCoFacs, ncol(corrMat))
        }
        transUnmixFlowObj <- arcTrans(
            flowObj = corrUnmixFlowObj,
            transNames = colnames(corrMat),
            transCoFacs = transCoFacs
        )
        return(transUnmixFlowObj)
    }
}