Added the functionality of scale.factor in NormalizeData being set to "median" of counts

R/preprocessing.R

@@ -3120,6 +3120,15 @@ RelativeCounts <- function(data, scale.factor = 1, verbose = TRUE) {
   if (verbose) {
     cat("Performing relative-counts-normalization\n", file = stderr())
   }
+
+  #setting scale.factor to be the median of counts across all columns if scale.factor is the string "median"
+  if (is.character(scale.factor) && scale.factor == "median") {
+    if(verbose){
+      cat("Calculating median scale factor\n", file = stderr())
+    }
+    scale.factor <- median(Matrix::colSums(data))
+  }
+
   norm.data <- data
   norm.data@x <- norm.data@x / rep.int(Matrix::colSums(norm.data), diff(norm.data@p)) * scale.factor
   return(norm.data)
@@ -4336,6 +4345,15 @@ LogNormalize.V3Matrix <- function(
   if (verbose) {
     cat("Performing log-normalization\n", file = stderr())
   }
+
+  #setting scale.factor to be the median of counts across all columns if scale.factor is the string "median"
+  if (is.character(scale.factor) && scale.factor == "median") {
+    if(verbose){
+      cat("Calculating median scale factor\n", file = stderr())
+    }
+    scale.factor <- median(Matrix::colSums(data))
+  }
+
   norm.data <- LogNorm(data, scale_factor = scale.factor, display_progress = verbose)
   colnames(x = norm.data) <- colnames(x = data)
   rownames(x = norm.data) <- rownames(x = data)

R/preprocessing5.R

@@ -256,6 +256,20 @@ LogNormalize.default <- function(
   if (isTRUE(x = verbose)) {
     pb <- txtProgressBar(file = stderr(), style = 3)
   }
+
+  #setting scale.factor to be the median of counts across all columns if scale.factor is the string "median"
+  if (is.character(scale.factor) && scale.factor == "median") {
+    if(verbose){
+      cat("Calculating median scale factor\n", file = stderr())
+    }
+    sums <- if (margin == 1L) {
+      rowSums(data)  # Sum of each row (gene) if margin is 1L
+    } else {
+      colSums(data)  # Sum of each column (cell) if margin is 2L
+    }
+    scale.factor = median(sums)
+  }
+
   for (i in seq_len(length.out = ncells)) {
     x <- if (margin == 1L) {
       data[i, ]
@@ -288,6 +302,15 @@ LogNormalize.IterableMatrix <- function(
     verbose = TRUE,
     ...
 ) {
+
+  #setting scale.factor to be the median of counts across all columns if scale.factor is the string "median"
+  if (is.character(scale.factor) && scale.factor == "median") {
+    if(verbose){
+      cat("Calculating median scale factor\n", file = stderr())
+    }
+    scale.factor <- median(colSums(data))
+  }
+
   data <- BPCells::t(BPCells::t(data) / colSums(data))
   # Log normalization
   data <- log1p(data * scale.factor)
@@ -860,6 +883,33 @@ DISP <- function(
     p <- p + 1L
   }
   np <- length(x = p) - 1L
+
+  #adding a progress bar for median calculation is verbose is TRUE
+  if (is.character(scale.factor) && scale.factor == "median" && isTRUE(x = verbose)) {
+    cat("Calculating median scale factor\n", file = stderr())
+    pb_median <- txtProgressBar(style = 3L, file = stderr())
+  }
+
+  #setting scale.factor to be the median of counts across all columns if scale.factor is the string "median"
+  if (is.character(scale.factor) && scale.factor == "median") {
+    col_sums <- numeric(np)
+    for (i in seq_len(length.out = np)) {
+      idx <- seq.int(from = p[i], to = p[i + 1] - 1L)
+      xidx <- slot(object = data, name = entryname)[idx]
+      col_sums[i] <- sum(xidx)
+
+      if (isTRUE(x = verbose)) {
+        setTxtProgressBar(pb_median, value = i / np)
+      }
+    }
+
+    if (isTRUE(x = verbose)) {
+      close(pb_median)
+    }
+
+    scale.factor <- median(col_sums)
+  }
+
   if (isTRUE(x = verbose)) {
     pb <- txtProgressBar(style = 3L, file = stderr())
   }

tests/testthat/test_preprocessing.R

@@ -97,6 +97,21 @@ test_that("Relative count normalization returns expected values", {
   expect_equal(rc.counts[2, 1], 14285.71, tolerance = 1e-6)
 })
 
+denseMatrix <- as.matrix(pbmc.test)  # Matrix to test LogNormalize.V3Matrix and RelativeCounts methods
+test_that("LogNormalize.V3Matrix computes median scale factor correctly", {
+  expectedMedian <- median(colSums(denseMatrix))
+  resultFromExpectedMedian <- LogNormalize.V3Matrix(data = denseMatrix, scale.factor = expectedMedian, margin = 2L, verbose = FALSE)
+  resultFromScaleFactorSetToMedian <- LogNormalize.V3Matrix(data = denseMatrix, scale.factor = "median", margin = 2L, verbose = FALSE)
+  expect_equal(as.matrix(resultFromExpectedMedian), as.matrix(resultFromScaleFactorSetToMedian), tolerance = 1e-6)
+})
+
+test_that("RelativeCounts computes median scale factor correctly", {
+  expectedMedian <- median(colSums(denseMatrix))
+  resultFromExpectedMedian <- RelativeCounts(data = denseMatrix, scale.factor = expectedMedian, verbose = FALSE)
+  resultFromScaleFactorSetToMedian <- RelativeCounts(data = denseMatrix, scale.factor = "median", verbose = FALSE)
+  expect_equal(as.matrix(resultFromExpectedMedian), as.matrix(resultFromScaleFactorSetToMedian), tolerance = 1e-6)
+})
+
 # Tests for v5 NormalizeData
 # --------------------------------------------------------------------------------
 context("v5 NormalizeData")
@@ -175,6 +190,44 @@ test_that("LogNormalize normalizes properly for BPCells", {
   )
 })
 
+test_that("LogNormalize.IterableMatrix computes median scale factor correctly", {
+  skip_on_cran()
+  library(Matrix)
+  skip_if_not_installed("BPCells")
+  library(BPCells)
+  mat_bpcells <- t(as(t(object[['RNA']]$counts ), "IterableMatrix"))
+  expectedMedian <- median(colSums(mat_bpcells))
+  resultFromExpectedMedian <- LogNormalize.IterableMatrix(data = mat_bpcells, scale.factor = expectedMedian, margin = 2L, verbose = FALSE)
+  resultFromScaleFactorSetToMedian <- LogNormalize.IterableMatrix(data = mat_bpcells, scale.factor = "median", margin = 2L, verbose = FALSE)
+  expect_equal(as.matrix(resultFromExpectedMedian), as.matrix(resultFromScaleFactorSetToMedian), tolerance = 1e-6)
+})
+
+denseMatrix <- as.matrix(pbmc.test)  # Matrix to test LogNormalize.default when scale.factor is set to "median"
+test_that("LogNormalize.default computes median scale factor correctly for both margin values", {
+  expectedMedianForMargin1L <- median(rowSums(denseMatrix))
+  expectedMedianForMargin2L <- median(colSums(denseMatrix))
+
+  resultFromExpectedMedianForMargin1L <- LogNormalize.default(data = denseMatrix, scale.factor = expectedMedianForMargin1L, margin = 1L, verbose = FALSE)
+  resultFromExpectedMedianForMargin2L <- LogNormalize.default(data = denseMatrix, scale.factor = expectedMedianForMargin2L, margin = 2L, verbose = FALSE)
+
+  resultsFromScaleFactorSetToMedianForMargin1L <- LogNormalize.default(data = denseMatrix, scale.factor = "median", margin = 1L, verbose = FALSE)#if the normalization is across rows (genes)
+  resultsFromScaleFactorSetToMedianForMargin2L <- LogNormalize.default(data = denseMatrix, scale.factor = "median", margin = 2L, verbose = FALSE)#if the normalization is across columns (cells)
+
+  expect_equal(as.matrix(resultFromExpectedMedianForMargin1L), as.matrix(resultsFromScaleFactorSetToMedianForMargin1L), tolerance = 1e-6)
+  expect_equal(as.matrix(resultFromExpectedMedianForMargin2L), as.matrix(resultsFromScaleFactorSetToMedianForMargin2L), tolerance = 1e-6)
+})
+
+theSparseMatrix <- as.sparse(denseMatrix) # Sparse Matrix to test .SparseNormalize computes median scale factor correctly
+test_that("LogNormalize.default computes median scale factor correctly for both margin values", {
+  expectedMedian <- median(colSums(theSparseMatrix))
+
+  resultFromExpectedMedian <- .SparseNormalize(data = theSparseMatrix, scale.factor = expectedMedian, verbose = FALSE)
+  resultsFromScaleFactorSetToMedian <- .SparseNormalize(data = theSparseMatrix, scale.factor = "median", verbose = FALSE)
+
+  expect_equal(resultFromExpectedMedian, resultsFromScaleFactorSetToMedian, tolerance = 1e-6)
+})
+
+
 # Tests for ScaleData
 # --------------------------------------------------------------------------------
 context("ScaleData")