R/F_residualPlot.R

Defines functions residualPlot

Documented in residualPlot

#' Make residual plots
#'
#' @param RCM an RCM object
#' @param Dim an integer, which dimension?
#' @param whichTaxa a character string or a character vector,
#'  for which taxa to plot the diagnostic plots
#' @param resid the type of residuals to use, either 'Deviance' or 'Pearson'
#' @param numTaxa an integer, the number of taxa to plot
#' @param mfrow passed on to par().
#' If not supplied will be calculated based on numTaxa
#' @param samColour,samShape Vectors or character strings denoting
#' the sample colour and shape respectively. If character string is provided,
#' the variables with this name is extracted from the phyloseq object in RCM
#' @param legendLabSize size of the legend labels
#' @param legendTitleSize size of the legend title
#' @param axisLabSize size of the axis labels
#' @param axisTitleSize size of the axis title
#' @param taxTitle A boolean, should taxon title be printed
#' @param h Position of reference line. Set to NA for no line
#'
#'@details If whichTaxa is 'run' or 'response' the taxa with the highest
#' run statistics or steepest slopes of the response function are plotted,
#'  numTax indicates the number. If whichTaxa is a character vector,
#'  these are interpreted as taxon names to plot.
#'  This function is mainly meant for linear response functions,
#'  but can be used for others too.
#'  The runs test statistic from the tseries package is used.
#'@return Plots a ggplot2-object to output
#'@export
#'@import ggplot2
#'@import phyloseq
#'@importFrom tseries runs.test
#'@seealso \code{\link{RCM}}
#'@examples
#'data(Zeller)
#' require(phyloseq)
#' tmpPhy = prune_taxa(taxa_names(Zeller)[1:120],
#' prune_samples(sample_names(Zeller)[1:75], Zeller))
#' #Subset for a quick fit
#' zellerRCMlin = RCM(tmpPhy, k = 2,
#' covariates = c('BMI','Age','Country','Diagnosis','Gender'),
#' responseFun = 'linear', round = TRUE, prevCutOff = 0.03)
#' residualPlot(zellerRCMlin)
residualPlot = function(RCM, Dim = 1, whichTaxa = "response",
    resid = "Deviance", numTaxa = 9, mfrow = NULL,
    samColour = NULL, samShape = NULL, legendLabSize = 15,
    legendTitleSize = 16, axisLabSize = 14, axisTitleSize = 16,
    taxTitle = TRUE, h = 0) {
    sampleScore = RCM$covariates %*% RCM$alpha[, Dim,
        drop = FALSE]
    if (resid == "Deviance") {
        resMat = getDevianceRes(RCM, seq_len(Dim))
    } else if (resid == "Pearson") {
        mu = extractE(RCM, seq_len(Dim))
        # Residuals are also based on lower dimensions
        thetaMat = matrix(byrow = TRUE, nrow = nrow(RCM$X),
            ncol = ncol(RCM$X), data = RCM$thetas[,
                switch(as.character(Dim), `0` = "Independence",
                    `0.5` = "Filtered",
                    paste0("Dim",Dim))])
        resMat = (RCM$X - mu)/sqrt(mu + mu^2/thetaMat)
    } else {
        stop("Unknown residual type!")
    }
    if (!whichTaxa %in% c("runs", "response")) {
        numTaxa = length(whichTaxa)
        idTaxa = whichTaxa
    } else {
        sizes = if (whichTaxa == "runs")
            apply(resMat > 0, 2, function(x) {
                runs.test(factor(x))$statistic
            }) else RCM$NB_params[2, , Dim]
        # Select taxa with longest runs or strongest
        # responses
        idTaxa = which(sizes >= sort(sizes, decreasing = TRUE)[numTaxa])
    }
    # Prepare the plotting facets
    mfrow = if (is.null(mfrow))
        rep(ceiling(sqrt(numTaxa)), 2) else mfrow
    parTmp = par(no.readonly = TRUE)
    on.exit(par(parTmp))
    par(mfrow = mfrow)
    resMat = resMat[, idTaxa, drop = FALSE]
    Colour = if (is.null(samColour))
      "black" else get_variable(RCM$physeq, samColour)
    Shape = if (is.null(samShape)){
      if(length(idTaxa)>1) 1 else "none"
    } else get_variable(RCM$physeq, samShape)
    if (length(idTaxa) > 1) {
        foo = lapply(colnames(resMat), function(tax) {
            plot(x = sampleScore, y = resMat[, tax],
                ylab = paste(resid, "residuals"),
              xlab = paste("Environmental score in dimension", Dim),
                main = tax, col = Colour, pch = Shape)
        })
        return(invisible())
    } else {
        Plot = ggplot(data.frame(x = c(sampleScore),
            y = c(resMat), samColour = Colour, samShape = Shape),
            mapping = aes_string(x = "x", y = "y",
                colour = "samColour", shape = "samShape")) +
            ylab(paste(resid, "residuals")) +
            xlab(paste("Environmental score in dimension",
            Dim)) + geom_point() + ggtitle(ifelse(taxTitle,
            colnames(resMat), ""))
        Plot = Plot + if (is.null(samShape))
            guides(shape = FALSE) else scale_shape_discrete(name = samShape)
        Plot = Plot + if (is.null(samColour)) {
            guides(colour = FALSE)
        } else if (is.factor(get_variable(RCM$physeq,
            samColour))) {
            scale_colour_discrete(name = samColour)
        } else scale_colour_continuous(name = samColour)
        Plot = Plot + geom_hline(yintercept = h, linetype = "dashed", col = "black")
        Plot = Plot + theme_bw() + theme(axis.title = element_text(size = axisTitleSize),
            axis.text = element_text(size = axisLabSize),
            legend.title = element_text(size = legendTitleSize),
            legend.text = element_text(size = legendLabSize))
        return(Plot)
    }
  }

Try the RCM package in your browser

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

RCM documentation built on Nov. 8, 2020, 5:22 p.m.