R/mbqnSimuData.R

In MBQN: Mean/Median-balanced quantile normalization

#' Generate a random/structured data matrix
#'
#' @description Generate a random data matrix with or without proteomics,
#' log-transformed feature intensity-like properties.
#' @param model character indicating one of the three different type of models:
#' \code{"rand"}(default) a Gaussian random matrix of size nrow x ncol
#' (default 1000 x 10), \code{"omics"} a Gaussian random matrix of
#' size 1264 x 18 that mimics intensity profiles and missing values as
#' present in real data, and \code{"omics.dep"} is the same as \code{"omics"}
#' but with an additional single, differentially expressed RI feature.
#' @param nrow number of rows of data matrix (only for \code{model = "rand"}).
#' @param ncol number of columns of data matrix
#' (only for \code{model = "rand"}).
# #' @param seed seed for random number generator.
#' @param show.fig logical inidicating whether data properties are plot to
#' figure (only for \code{model = "omics"} and \code{model = "omics.dep"}).
#' @details For model \code{"rand"}, each matrix element is drawn from a
#' standard normal distribution \eqn{N(0,1)}. For model \code{"omics"}, the
#' matrix elements of each row are drawn from a Gaussian distribution
#' \eqn{N(\mu_i,\sigma_i^2)} where the mean and standard deviation itself are
#' drawn Gaussian distributions, i.e. \eqn{\sigma_i~N(0,0.0625)} and
#' \eqn{\mu_i~N(28,4)}. About 35\% of the matrix values are set to NA according
#' to the missing value pattern present in real protein LFQ
#' intensities. For model \code{"omics.dep"}, a single differentially epxressed
#' RI feature is stacked on top of the matrix from model \code{"omics"}.
#' @return \code{matrix} of size nrow x ncol.
#' @seealso [example_NApattern()] for description of missing value pattern.
#' @importFrom utils read.csv untar unzip
#' @importFrom stats rnorm
#' @importFrom graphics image layout points rect
#' @references Brombacher, E., Schad, A., Kreutz, C. (2020). Tail-Robust
#' Quantile Normalization. BioRxiv.
#' @examples
#' mbqnSimuData(model = "rand")
#' mbqnSimuData(model = "rand", 2000,6)
#' set.seed(1234)
#' mbqnSimuData(model = "omics")
#' set.seed(1111)
#' mbqnSimuData(model = "omics.dep")
#' @author Ariane Schad
#' @export mbqnSimuData
mbqnSimuData <- function(model = "rand", nrow = NULL, ncol = NULL,
                        show.fig = FALSE){

    if (is.null(nrow)) nrow <- 1000
    if (is.null(ncol)) ncol <- 10

    if (model=="rand"){
        # generate a random matrix without NAs
        dat <- replicate(ncol, rnorm(nrow))
        if(show.fig){
            image(t(dat), xlab = "sample", ylab = "feature row",
                main = "simulated data", axes= FALSE)
            axis(1, at = seq_len(ncol(dat))/ncol(dat),
                labels = seq_len(ncol(dat)))
        }
    }

    if (model == "omics" || model == "omics.dep"){
        # generate a structured random matrix with NAs
        # the NA structure is extracted from a real dataset

        # load an internal stored MV pattern extracted from a real proteomics
        # dataset from PRIDE
        mtx <- example_NApattern
        mtx[mtx==0] <- NA

        dat <- replicate(
            ncol(mtx), rnorm(nrow(mtx))*0.25)+rnorm(nrow(mtx))*2+28
        s.dat <-sort(apply(dat, 1,mean, na.rm =TRUE),
                        index.return = TRUE,
                        decreasing = TRUE)
        dat <- dat[s.dat$ix,]
        dat[is.na(mtx)] <- NA

        if (show.fig){
            par(mfrow=c(2,2))
            image(t(mtx), xlab = "sample", ylab = "feature row (sorted)",
                    main = "MV pattern", axes = FALSE)
            axis(1, at = seq_len(ncol(dat))/ncol(dat),
                    labels = seq_len(ncol(dat)))
            image(t(dat), xlab = "sample", main = "simulated", axes= FALSE)
            axis(1, at = seq_len(ncol(dat))/ncol(dat),
                    labels = seq_len(ncol(dat)))

            plot(apply(dat,1,mean,na.rm =TRUE),apply(is.na(dat),1,
                        mean,na.rm =TRUE),
                col =1,
                xlab = "mean feature intensity", ylab = "MV frequency")
            legend("topright",legend = c("simulated"),  pch = 1, col = c(1),
                bty ="n", y.intersp = 1.5)
        }

        if (model == "omics.dep"){
            # add extra feature with diff. expr.
            ncol <- ncol(dat)
            dat <- rbind(c(rep(34.7,9),rep(34.6, 9))+rnorm(ncol)*0.02,dat)
        }
    }
    return(dat)
}