R/plotExpr.R

In WaltersLab/doseR: doseR

#' @title plotExpr Function to generate a boxplot (expression) for a
#' SummarizedExperiment object based on the replicate data.
#' @description This function generates a boxplot of FPKM expression
#' values from the supplied object. FPKM values are averaged
#' across replicates and partitioned among groups of loci as specified
#' in a selected column from the annotation slot of the provided
#' object.
#' @usage plotExpr(se, groupings= NULL, mode_mean=TRUE,
#' treatment=levels(colData(se)$Treatment),
#' LOG2=TRUE, clusterby_grouping=TRUE, ...)
#' @param se A SummarizedExp object containing FPKM values and at least
#' one annotation column.
#' @param groupings Specifies which column in the dataframe of the
#' annotation slot that will be used to group loci in the boxplot.
#' Can provide either a character value matching the column name,
#' or a single numerical value used as an index of dataframe columns.
#' @param mode_mean Logical. If TRUE then FPKM values are averaged by
#' mean across replicates within treatment. If FALSE, values are
#' averaged by median.
#' @param treatment A character vector indicating which treatments
#' (i.e. levels in the replicates slot vector) will be plotted.
#' Order matters, and controls the ordering of treatments represented
#' in the boxplot.
#' @param LOG2 Logical. If TRUE then average FPKM values are Log2
#' transformed.
#' @param clusterby_grouping Logical. If TRUE then boxplots are arranged
#' by locus annotation grouping. If FALSE they are arranged by treatment
#' levels, as indicated in the treatment argument.
#' @param ... Additional named arguments and graphical parameters passed
#' to the boxplot function.
#' @details This function generates boxplots to visualize the distribution
#' of FPKM expression values provided in an object, arranged by
#' selected treatments and locus annotations. FPKM values are averaged
#' (mean or median) within selected treatments, to provide a single
#' expression value per locus per treatment. Loci are partitioned
#' into groupings based on a specified column in the dataframe of
#' annotations slot of the object. Thus a box is drawn for
#' each grouping of loci for each treatment indicated. Desired treatments
#' and their ordering are specified by the treatment argument. Groupings
#' are arranged by sort order of the annotation column indicated, and can
#' thus be controlled by providing a factor with a pre-specified level order.
#' By default (clusterby_grouping = TRUE), boxes are arranged by annotation
#' group first, and then by treatment, but setting this option to FALSE
#' arranges boxes by treatment and then annotation group. This function uses
#' the base graphics boxplot function to generate the plot, so can accept
#' all relevant graphical arguments for customizing the figure;
#' see boxplot for details.
#' @examples
#' data(hmel.se)
#' plotExpr(se, groupings = "annotation.ZA", treatment = 'Male' )
#' @return Returns an invisible data frame containing values
#' and labels used to generate the figure.
#' @author AJ Vaestermark, JR Walters.
#' @references The 'doseR' package, 2018 (in press).

plotExpr <- function(se, groupings = NULL, mode_mean = TRUE,
treatment=levels(colData(se)$Treatment), LOG2 = TRUE, clusterby_grouping=TRUE,
...) {
    MyGroups <- rowData(se)[[groupings]]
    if (is.null(groupings)) {
        stop("No groupings, e.g. groupings=\"something\"..."); return(NULL)
    }
    if (is.element(FALSE, treatment %in% levels(colData(se)$Treatment))) {
stop("Some treatment not in levels(colData(se)$Treatment), please check...")
        return(NULL)
    }
    MyLabels <- NULL ; PLOT <- NULL ; NAMES <- NULL
    if (is.factor(MyGroups)) {
        MyGroups <- droplevels(MyGroups)
        Super_ch <- if (clusterby_grouping)
            levels(MyGroups) else treatment
        Super_dh <- if (clusterby_grouping)
            treatment else levels(MyGroups)
    } else {
        Super_ch <- if (clusterby_grouping) unique(MyGroups) else treatment
        Super_dh <- if (clusterby_grouping) treatment else unique(MyGroups)
    }
    if (is.factor(colData(se)$Treatment)) {
colData(se)$Treatment <- droplevels(colData(se)$Treatment)
    }
    for (ch in Super_ch) {
        for (dh in Super_dh) {
            actual_ch <- if (clusterby_grouping)
                dh else ch
            actual_dh <- if (clusterby_grouping)
                ch else dh
            if (is.element(actual_ch, treatment)) {
                NAMES <- c(NAMES, paste0(actual_ch, actual_dh))
                RM <- if (mode_mean)
rowMeans(assays(se)$rpkm[, colData(se)$Treatment == actual_ch]) else
matrixStats::rowMedians(assays(se)$rpkm[, colData(se)$Treatment == actual_ch])
                PLOT <- c(PLOT, RM[MyGroups == actual_dh])
                MyLabels <- c(MyLabels, rep(paste0(actual_ch, actual_dh),
length(RM[MyGroups == actual_dh])))

            }
        }
    }
    if (LOG2) { PLOT <- log2(PLOT) }
    PLOT[is.infinite(PLOT)] <- NA
    MyLabels <- factor(MyLabels, levels = NAMES)
    boxplot(PLOT ~ MyLabels, ...)
    invisible(data.frame(values = PLOT, labels = MyLabels))
}  # plotExpr