R/qc.R

Defines functions densityPlot densityBeanPlot controlStripPlot qcReport

Documented in controlStripPlot densityBeanPlot densityPlot qcReport

# Exported functions -----------------------------------------------------------

qcReport <- function(rgSet, sampNames = NULL, sampGroups = NULL,
                     pdf = "qcReport.pdf", maxSamplesPerPage = 24,
                     controls = c("BISULFITE CONVERSION I",
                                  "BISULFITE CONVERSION II", "EXTENSION",
                                  "HYBRIDIZATION", "NON-POLYMORPHIC",
                                  "SPECIFICITY I", "SPECIFICITY II",
                                  "TARGET REMOVAL")) {
    .isRGOrStop(rgSet)
    if (is.null(sampNames)) sampNames <- colnames(rgSet)
    n <- ncol(rgSet)
    o <- rev(order(sampNames))
    rgSet <- rgSet[, o]
    sampNames <- sampNames[o]
    if (is.null(sampGroups)) sampGroups <- rep(1, n)
    sampGroups <- sampGroups[o]
    sampGroups <- as.factor(sampGroups)
    numPages <- ceiling(n/maxSamplesPerPage)
    samplesPerPage <- ceiling(n/numPages)
    sampleIdxs <- suppressWarnings(
        split(seq_len(n), rep(seq_len(numPages), each = samplesPerPage)))
    pdf(file = pdf, width = 8, height = 11)
    par(mfrow = c(2, 1))
    densityPlot(rgSet, sampGroups = sampGroups, main = "Beta", xlab = "Beta")
    plot.new()
    par(mfrow = c(1, 1), oma = c(2, 10, 1, 1))
    for (sampleIdx in sampleIdxs) {
        densityBeanPlot(
            dat = rgSet[, sampleIdx],
            sampGroups = sampGroups[sampleIdx],
            sampNames = sampNames[sampleIdx])
    }
    for (controlType in controls) {
        for (sampleIdx in sampleIdxs) {
            controlStripPlot(
                rgSet = rgSet[, sampleIdx],
                sampNames = sampNames[sampleIdx],
                controls = controlType)
        }
    }
    dev.off()
}

controlStripPlot <- function(rgSet,
                             controls = c("BISULFITE CONVERSION I",
                                          "BISULFITE CONVERSION II"),
                             sampNames = NULL, xlim = c(5, 17)) {
    .isRGOrStop(rgSet)

    r <- getRed(rgSet)
    g <- getGreen(rgSet)

    for (controlType in controls) {
        ctrlAddress <- getControlAddress(rgSet, controlType = controlType)

        # Red channel
        ctlWide <- as.matrix(log2(r[ctrlAddress, ,drop = FALSE]))
        if (!is.null(sampNames)) colnames(ctlWide) <- sampNames
        ctlR <- melt(ctlWide, varnames = c("address", "sample"))

        # Green channel
        ctlWide <- as.matrix(log2(g[ctrlAddress, ,drop = FALSE]))
        if (!is.null(sampNames)) colnames(ctlWide) <- sampNames
        ctlG <- melt(ctlWide, varnames = c("address", "sample"))

        # Plot
        ctl <- rbind(
            cbind(channel = "Red", ctlR),
            cbind(channel = "Green", ctlG))
        if (any((ctl$value < xlim[1]) | (ctl$value > xlim[2]))) {
            message("Warning: ", controlType, " probes outside plot range")
        }
        fig <- xyplot(
            x = sample ~ value | channel,
            groups = channel, horizontal = TRUE, pch = 19,
            col = c("darkred", "darkgreen"),
            xlab = "Log2 Intensity",
            xlim = xlim,
            main = paste("Control:", controlType),
            layout = c(2, 1),
            data = ctl,
            panel = function(x, y,...) {
                panel.stripplot(x, y,...)
                panel.abline(h = (as.numeric(y) - 0.5), lty = 3, col = "grey70")
            })
        print(fig)
    }
}

densityBeanPlot <- function(dat, sampGroups = NULL, sampNames = NULL,
                            main = NULL, pal = brewer.pal(8, "Dark2"),
                            numPositions = 10000) {
    if (is(dat, "RGChannelSet") || is(dat, "MethylSet")) {
        b <- getBeta(dat)
    } else if (is(dat, "matrix") || is(dat, "DelayedMatrix")) {
        b <- dat
    } else {
        stop("argument 'dat' must be an 'RGChannelSet', a 'MethylSet', ",
             "a 'matrix', or a 'DelayedMatrix'.")
    }
    n <- ncol(b)
    if (!is.null(sampNames)) colnames(b) <- sampNames
    if (is.null(main)) main <- "Beta"
    if (is.null(sampGroups))  sampGroups <- rep(1, n)
    sampGroups <- as.factor(sampGroups)
    col <- lapply(sampGroups, function(x) rep(pal[x],4))
    if (is.null(numPositions)) {
        idx <- 1:dim(dat)[1]
    } else {
        idx <- sample(nrow(b), numPositions)
    }
    # NOTE: At this point, the data need to be in-memory as an ordinary matrix
    # TODO: Update in response to any suggestions in
    #       https://github.com/Bioconductor/DelayedArray/issues/13
    b_subset <- as.matrix(b[idx, ])
    x <- melt(b_subset, varnames = c("cpg", "sample"))
    o <- order(colnames(b))
    beanplot(
        value ~ sample,
        horizontal = TRUE,
        what = c(0, 1, 1, 0),
        log = "",
        las = 1,
        ylim = c(0, 1),
        xlab = "Beta",
        main = main,
        col = col[o],
        data = x,
        cex.lab = 0.9,
        beanlinewd = 1,
        border = NA)
    abline(h = seq(1, n + 1) - 0.5, lty = 3, col = "grey70")
}


densityPlot <- function(dat, sampGroups = NULL, main = "", xlab = "Beta",
                        pal = brewer.pal(8, "Dark2"),
                        xlim, ylim, add = TRUE, legend = TRUE, ...) {
    if (is(dat, "RGChannelSet") || is(dat, "MethylSet")) {
        b <- getBeta(dat)
    } else if (is(dat, "matrix")) {
        b <- dat
    } else {
        stop("argument 'dat' must be an 'RGChannelSet', a 'MethylSet' or ",
             "matrix.")
    }
    # NOTE: Have to ensure data are in memory as an ordinary vector before
    #       computing density
    d <- apply(b, 2, function(x) density(as.vector(x), na.rm = TRUE))
    if (missing(ylim)) ylim <- range(sapply(d, function(i) range(i$y)))
    if (missing(xlim)) xlim <- range(sapply(d, function(i) range(i$x)))
    if (is.null(sampGroups)) {
        sampGroups <- rep(1, ncol(b))
    } else if (length(sampGroups) == 1) {
        sampGroups <- rep(sampGroups, ncol(b))
    }
    sampGroups <- as.factor(sampGroups)
    if (add) {
        plot(x = 0,
             type = "n",
             ylim = ylim,
             xlim = xlim,
             ylab = "Density",
             xlab = xlab,
             main = main, ...)
        abline(h = 0, col = "grey80")
    }
    for (i in seq_along(d)) {
        lines(d[[i]], col = pal[sampGroups[i]])
    }
    if (legend & length(levels(sampGroups)) > 1) {
        legend("topright", legend = levels(sampGroups), text.col = pal)
    }
}
kasperdanielhansen/minfi documentation built on May 4, 2024, 12:04 p.m.