R/do_variabilty_list.R

In computational-metabolomics/qcrms: Creates quality control output from XCMS output

#' @importFrom stats sd mad
NULL

#' Calculate RSD\% or MAD values per sample group of input peakmatrix
#'
#' @param peak_data peak matrix
#' @param classes Vector of class labels
#' @param method What method to use to summarise values. Supported methods are
#'  'RSD', 'MAD', 'median' and 'none'.
#' @return List of RSD\% or MAD values for each feature for each sample group
#' @export

do_variability_list <- function(peak_data, classes, method="RSD") {

  peak_data <- pmp:::check_input_data(df=peak_data, classes=classes)

  cl <- unique(classes)
  out <- vector("list", length(cl))
  names(out) <- cl

  #Calculate RSD before scaling and MV imputation
  if (method=="RSD") {
    FUN <- function(x) sd(x, na.rm=TRUE) / mean(x, na.rm=TRUE) * 100.0
  } else if (method == "MAD") {
    FUN <- function(x) mad(x, na.rm=TRUE, constant=1.4826)
  } else if (method == "median") {
    FUN <- function(x) median(x, na.rm=TRUE)
  } else if (method == "none") {
    FUN <- NULL
  }else {
    stop("Method specified is not supported.")
  }

  for (slab in seq_len(length(out))) {
    if (method=="none") {
      out[[slab]] <-
        c(as.matrix(assay(peak_data)[, classes==names(out)[slab]]))
    } else {
      out[[slab]] <- apply(rbind(assay(peak_data)[, classes==names(out)[slab]],
        NULL), 1L, FUN)
    }
  }
  out$variability_method <- method
  out
}