R/oneVsAllPlot.R

Defines functions oneVsAllPlotCoFunction oneVsAllPlot

Documented in oneVsAllPlot

#' Plotting all variables against a single variable
#'
#'
#' This function is useful both when setting appropriate gates and when the
#' adjustments of the compensation are done
#' @param flowObj This is the full dataset, either a flowFrame or a flowSet,
#' that should be plotted. If it has more rows than "nRows", a subsample (with
#' equal contributions from each sample if a flowSet) will be plotted.
#' @param yCol Here, the variable to be plotted against all the others is
#' selected. It can be either a number, the column name of interest or "all".
#' @param nRows The number of rows that will be used to construct the plot.
#' The fewer, the faster, but the resolution also decreases, naturally. Default
#' is 100000.
#' @param plotName If a name different from yCol should be used, it can be added
#' here.
#' @param zeroTrim In the case of CyTOF data, the events at zero can often
#' be so dominant, that all other density variation is dwarfed, and thus
#' invisible. With this command, the events that are zero in both y
#' and x are trimmed for each x separately.
#' @param saveResult Should the result be saved as a file?
#' @return A plot with one 2D-graph for each variable that the y-variable
#' should be plotted against.
#' @seealso \code{\link{correctUnmix}}
#' @importFrom flowCore fsApply exprs
#' @importFrom reshape2 melt
#' @importFrom ggplot2 ggplot aes facet_wrap geom_hex xlab ylab theme
#' element_blank element_line ggsave scale_fill_gradientn
#' @importFrom BiocParallel bplapply
#' @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)
#'
#' # Now unmix
#' fullPanelUnmix <- specUnmix(fullPanel, specMat)
#'
#' # Transform all fluorescent channels
#' fullPanelTrans <- arcTrans(fullPanelUnmix,
#'     transNames = colnames(fullPanelUnmix)[6:18])
#'
#' # And now run the function. If no specific marker is selected, as in this
#' # case, then all markers will be plotted in a new sub-directory.
#' # Further, if you leave the saveResult to TRUE, a pdf will be created.
#' oneVsAllPlot(fullPanelTrans, yCol = "BV650_CD56", saveResult = FALSE)
#'
#' # This shows that there is a compensation artifact between AF647_IgM and
#' # BV650_CD56, which is an expected combination to cause problems, due to the
#' # similar emission characteristics. It is therefore recommended to go on to
#' # correctUnmix function.
#'
#' @export oneVsAllPlot
oneVsAllPlot <- function(flowObj, yCol = "all", nRows = 10000,
                         plotName = "default",
                         zeroTrim = TRUE, saveResult = TRUE) {
    plotExprs <- plotDownSample(flowObj, nRows)

    if (yCol == "all") {
        dir.create("All_vs_all_plots")
        bplapply(seq_along(colnames(plotExprs)), function(x) {
            oneVsAllPlotCoFunction(
                plotExprs = plotExprs, yCol = x,
                plotName = plotName,
                zeroTrim = zeroTrim,
                saveResult = saveResult,
                dirName = "All_vs_all_plots"
            )
        })
    } else {
        oneVsAllPlotCoFunction(
            plotExprs = plotExprs, yCol = yCol,
            plotName = plotName,
            zeroTrim = zeroTrim,
            saveResult = saveResult
        )
    }
}

oneVsAllPlotCoFunction <- function(plotExprs, yCol, plotName, zeroTrim,
                                   saveResult, dirName = ".") {
    # If the yCol is given as a character, it is converted to the correct number
    # here
    if (is.character(yCol)) {
        yCol <- which(colnames(plotExprs) == yCol)
    }

    if (plotName == "default") {
        plotName <- file.path(dirName, colnames(plotExprs)[yCol])
    }

    # Here, a separate y-name is created
    yName <- colnames(plotExprs)[yCol]

    # Now, the data is truncated, to minimize influence by individual events
    # on the display
    plotExprs <- apply(plotExprs, 2, function(x) {
        high <- quantile(x, 0.999)
        low <- quantile(x, 0.001)
        x[x > high] <- high
        x[x < low] <- low
        return(x)
    })

    # Here, the dataset is converted into a long format.
    longPlot <- melt(as.data.frame(plotExprs))
    longPlot$Y <- rep(plotExprs[, yCol], times = ncol(plotExprs))

    # Here, the double negative events are made NA for each x separately
    if (zeroTrim == TRUE) {
        longPlot[which(longPlot$value == 0 & longPlot$Y == 0), 2:3] <- c(NA, NA)
    }

    # Now, the plots are created.
    # Before this, we are just formally defining vale and Y, for the sake of
    # notes in CMD check.
    value <- 0
    Y <- 0
    p <- ggplot(longPlot, aes(x = value, y = Y)) +
        facet_wrap(~variable, scales = "free") +
        geom_hex(na.rm = TRUE) +
        scale_fill_gradientn(colours = c(
            "#440154FF", "#38598CFF", "#2D708EFF",
            "#25858EFF", "#1E9B8AFF", "#2BB07FFF",
            "#51C56AFF", "#85D54AFF", "#C2DF23FF",
            "#FDE725FF", "#FDE725FF", "#FDE725FF"
        )) +
        xlab("") + ylab(yName) + theme(
            axis.text = element_blank(),
            axis.ticks = element_blank(),
            panel.grid.major = element_blank(),
            panel.grid.minor = element_blank(),
            panel.background = element_blank(),
            axis.line = element_line(colour = "black")
        )
    if (saveResult) {
        ggsave(paste0(plotName, "_vs_all_others.pdf"))
    } else {
        print(p)
    }
}

Try the flowSpecs package in your browser

Any scripts or data that you put into this service are public.

flowSpecs documentation built on Nov. 8, 2020, 5:39 p.m.