R/abis.R

In grst/immunedeconv: Methods for immune cell deconvolution

#' Use ABIS to deconvolute a gene expression matrix.
#'
#' Source code adapted from https://github.com/giannimonaco/ABIS -
#'   server.R of the Shiny application
#' @param expression numeric matrix; genes as rows and patients as columns.
#'     Must be TP normalized
#' @param array logical; specifies whether the expression data are from
#'     microarrays (TRUE) or RNAseq (FALSE)
#'
#' @importFrom MASS rlm
deconvolute_abis_default <- function(expression, array = FALSE) {
  # 2 sets of operations for RNA/Microarray

  signature.path <- system.file("extdata", "abis", package = "immunedeconv", mustWork = TRUE)

  if (array == FALSE) {
    signature.path.rna <- paste0(signature.path, "/sigmatrixRNAseq.txt")
    signature <- read.delim(signature.path.rna, check.names = F)
    genes <- intersect(rownames(expression), rownames(signature))
    Dec <- (apply(expression[genes, , drop = F], 2, function(x) coef(rlm(as.matrix(signature[genes, ]), x, maxit = 100)))) * 100
  } else {
    signature.path.micro <- paste0(signature.path, "/sigmatrixMicro.txt")
    signature.path.target <- paste0(signature.path, "/target.txt")

    signature <- read.delim(signature.path.micro, check.names = F)
    target <- read.delim(signature.path.target, header = F)
    genes <- intersect(rownames(expression), rownames(signature))
    expression2 <- preprocessCore::normalize.quantiles.use.target(as.matrix(expression[genes, ]), target[, 1], copy = TRUE, subset = NULL)
    colnames(expression2) <- colnames(expression)
    rownames(expression2) <- genes
    Dec <- (apply(expression2, 2, function(x) coef(rlm(as.matrix(signature[genes, ]), x, maxit = 100)))) * 100
  }

  Dec <- signif(Dec, 3)
  # Dec[Dec < 0] <- 0

  Dec
}