Add log_transform parameter to deconvolute_mcp_counter

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: immunedeconv
 Title: Methods for immune cell deconvolution
 Description: Collection of methods for immune cell deconvolution from bulk RNA-seq samples.
-Version: 2.1.0
+Version: 2.1.4
 Authors@R: c(
     person("Lorenzo", "Merotto", role=c("aut", "cre"), email="lorenzo.merotto@uibk.ac.at"),
     person("Gregor", "Sturm", role=c("aut"), email="mail@gregor-sturm.de")
@@ -66,4 +66,4 @@ Roxygen: list(markdown = TRUE)
 LazyData: true
 URL: https:/omnideconv.org/immunedeconv, https://github.com/omnideconv/immunedeconv
 BugReports: https://github.com/omnideconv/immunedeconv/issues
-RoxygenNote: 7.3.3
+Config/roxygen2/version: 8.0.0

R/cell_type_mapping.R

@@ -134,7 +134,7 @@ summarise_children <- function(df) {
   df[set_to_zero] <- 0
 
   # finally, sum up the children.
-  df_sum <- summarise_all(df, funs(sum))
+  df_sum <- summarise(df, across(everything(), sum))
   df_sum
 }
 

R/immune_deconvolution_methods.R

@@ -159,12 +159,43 @@ deconvolute_xcell <- function(gene_expression_matrix, arrays, expected_cell_type
 #' @param gene_expression_matrix a m x n matrix with m genes and n samples
 #' @param feature_types type of identifiers used for expression features. May be
 #'  one of `"affy133P2_probesets","HUGO_symbols","ENTREZ_ID"`
+#' @param log_transform  Controls whether the expression matrix is log2-transformed before
+#'  running MCP-counter. MCP-counter expects log-transformed data. One of `NULL` (default),
+#'  `TRUE`, or `FALSE`.
+#'  \itemize{
+#'    \item `NULL` – auto-detect: if `max(gene_expression_matrix) > 50` the data are assumed
+#'      to be in linear (TPM) scale and will be log2(x + 1)-transformed.
+#'    \item `TRUE` – always apply log2(x + 1) transformation.
+#'    \item `FALSE` – assume data are already log-transformed; no transformation is applied.
+#'  }
 #' @param ... passed through to original MCP-counter function. A native argument takes precedence
 #'   over an immunedeconv argument (e.g. `featureType` takes precedence over `feature_types`)
 #'  See [MCPcounter.estimate](https://github.com/ebecht/MCPcounter/blob/master/Source/R/MCPcounter.R#L19).
 #'
 #' @export
-deconvolute_mcp_counter <- function(gene_expression_matrix, feature_types = "HUGO_symbols", ...) {
+deconvolute_mcp_counter <- function(gene_expression_matrix, feature_types = "HUGO_symbols",
+                                    log_transform = NULL, ...) {
+  if (is.null(log_transform)) {
+    if (max(gene_expression_matrix, na.rm = TRUE) > 50) {
+      message(
+        "MCP-counter expects log-transformed expression data. ",
+        "The input data appears to be in linear (TPM) scale (max value > 50) and will be ",
+        "log2(x + 1)-transformed automatically. Set `log_transform = TRUE` or ",
+        "`log_transform = FALSE` to force or disable this behaviour."
+      )
+      gene_expression_matrix <- log2(as.matrix(gene_expression_matrix) + 1)
+    } else {
+      message(
+        "MCP-counter expects log-transformed expression data. ",
+        "The input data appears to be already log-transformed (max value <= 50). ",
+        "No transformation applied. Set `log_transform = TRUE` or ",
+        "`log_transform = FALSE` to override this automatic detection."
+      )
+    }
+  } else if (isTRUE(log_transform)) {
+    gene_expression_matrix <- log2(as.matrix(gene_expression_matrix) + 1)
+  }
+
   arguments <- dots_list(gene_expression_matrix, featuresType = feature_types, ..., .homonyms = "last")
   call <- rlang::call2(MCPcounter::MCPcounter.estimate, !!!arguments)
   eval(call)

tests/testthat/test_deconvolution.R

@@ -41,6 +41,45 @@ test_that("mcp_counter works", {
   assert("matrix dimensions consistent", ncol(res) == ncol(test_mat))
 })
 
+test_that("mcp_counter log_transform auto-detects raw TPM data", {
+  # test_mat has max value > 50, so auto-detection should log-transform
+  expect_message(
+    res <- deconvolute_mcp_counter(test_mat),
+    "log2\\(x \\+ 1\\)-transformed automatically"
+  )
+  assert("matrix dimensions consistent", ncol(res) == ncol(test_mat))
+})
+
+test_that("mcp_counter log_transform auto-detects already log-transformed data", {
+  log_mat <- log2(test_mat + 1)
+  # max value should be <= 50, so auto-detection should not transform
+  expect_message(
+    res <- deconvolute_mcp_counter(log_mat),
+    "appears to be already log-transformed"
+  )
+  assert("matrix dimensions consistent", ncol(res) == ncol(log_mat))
+})
+
+test_that("mcp_counter log_transform NULL and TRUE give same result for raw TPM", {
+  # test_mat is raw TPM (max > 50), so NULL should auto-transform just like TRUE
+  suppressMessages(res_null <- deconvolute_mcp_counter(test_mat, log_transform = NULL))
+  res_true <- deconvolute_mcp_counter(test_mat, log_transform = TRUE)
+  expect_equal(res_null, res_true)
+})
+
+test_that("mcp_counter log_transform NULL and FALSE give same result for log-transformed data", {
+  log_mat <- log2(test_mat + 1)
+  # log_mat is already log-transformed (max <= 50), so NULL should skip like FALSE
+  suppressMessages(res_null <- deconvolute_mcp_counter(log_mat, log_transform = NULL))
+  res_false <- deconvolute_mcp_counter(log_mat, log_transform = FALSE)
+  expect_equal(res_null, res_false)
+})
+
+test_that("mcp_counter log_transform can be passed via deconvolute", {
+  res <- deconvolute(test_mat, "mcp_counter", log_transform = FALSE)
+  assert("matrix dimensions consistent", ncol(res) == ncol(test_mat) + 1)
+})
+
 test_that("epic works", {
   res <- deconvolute_epic(test_mat, tumor = TRUE, scale_mrna = TRUE)
   assert("matrix dimensions consistent", ncol(res) == ncol(test_mat))