R/CellEntropy.R

In zh542370159/dropSplit: Automatically identify cell-containing and empty droplets for droplet-based scRNAseq data using dropSplit

#' Calculate entropy for each cell according to the feature counts.
#'
#' @param x A sparse matrix which columns represent cells and rows represent features.
#'
#' @return A vector of the cell entropy.
#'
#' @examples
#' x <- simSimpleCounts()
#' CE <- CellEntropy(x)
#' head(CE)
#' @importFrom Matrix t colSums
#' @export
CellEntropy <- function(x) {
  if (!class(x) %in% c("dgCMatrix", "dgTMatrix")) {
    stop("'x' must be sparse Matrix of class dgCMatrix or dgTMatrix")
  }
  logp_matrix <- p_matrix <- Matrix::t(Matrix::t(x) / Matrix::colSums(x))
  logp_matrix@x <- log2(logp_matrix@x)
  e <- p_matrix * logp_matrix
  entropy <- -Matrix::colSums(e, na.rm = TRUE)
  names(entropy) <- colnames(x)
  return(entropy)
}


#' Calculate the max entropy for the cells when given the same counts.
#'
#' @param x A sparse matrix which columns represent cells and rows represent features.
#'
#' @return A vector of the max cell entropy.
#'
#' @examples
#' x <- simSimpleCounts()
#' maxCE <- maxCellEntropy(x)
#' head(maxCE)
#' @importFrom Matrix t colSums
#' @export
maxCellEntropy <- function(x) {
  if (!class(x) %in% c("dgCMatrix", "dgTMatrix")) {
    stop("'x' must be sparse Matrix of class dgCMatrix or dgTMatrix")
  }
  nCount <- Matrix::colSums(x)
  nGene <- Matrix::colSums(x > 0)
  i <- nCount %/% nGene
  r <- nCount %% nGene
  pi <- i / nCount
  pr <- (i + 1) / nCount
  ei <- (nGene - r) * pi * log2(pi)
  er <- r * pr * log2(pr)
  ei[is.na(ei)] <- 0
  er[is.na(er)] <- 0
  entropy <- -(ei + er)
  names(entropy) <- colnames(x)
  return(entropy)
}

#' Calculate the Efficiency for the cells according to the feature counts.
#'
#' @param x A sparse matrix which columns represent cells and rows represent features.
#'
#' @return A vector of the CellEfficiency.
#'
#' @examples
#' x <- simSimpleCounts()
#' ce <- CellEfficiency(x)
#' head(ce)
#' @export
CellEfficiency <- function(x) {
  if (!class(x) %in% c("dgCMatrix", "dgTMatrix")) {
    stop("'x' must be sparse Matrix of class dgCMatrix or dgTMatrix")
  }
  e <- CellEntropy(x) / maxCellEntropy(x)
  e[is.na(e)] <- 1
  names(e) <- colnames(x)
  return(e)
}

#' Calculate the Redundancy for the cells according to the feature counts.
#'
#' @param x A sparse matrix which columns represent cells and rows represent features.
#'
#' @return A vector of the CellRedundancy.
#'
#' @examples
#' x <- simSimpleCounts()
#' cr <- CellRedundancy(x)
#' head(cr)
#' @export
CellRedundancy <- function(x) {
  if (!class(x) %in% c("dgCMatrix", "dgTMatrix")) {
    stop("'x' must be sparse Matrix of class dgCMatrix or dgTMatrix")
  }
  r <- 1 - CellEfficiency(x)
  r[r < 0] <- 0
  names(r) <- colnames(x)
  return(r)
}