Fix normalization checks

CHANGELOG.md

@@ -16,6 +16,10 @@
 
 * `workflows/differential_expression/pseudobulk_deseq2`: Workflow for generating pseudobulk samples from single-cell data followed by DESeq2 differential expression analysis (PR #1044)
 
+## BUG FIX
+
+* `differential_expression/create_pseudobulks`: Fixed the check to verify that the raw counts layer was passed (PR #1072).
+
 # openpipelines 3.0.0
 
 ## BREAKING CHANGES

src/differential_expression/create_pseudobulk/script.py

@@ -27,12 +27,21 @@
 
 
 def is_normalized(layer):
+    exp_layer = np.expm1(layer)  # Inverse of log1p
+
     if sp.issparse(layer):
         row_sums = np.array(layer.sum(axis=1)).flatten()
+        exp_row_sums = np.array(exp_layer.sum(axis=1)).flatten()
     else:
         row_sums = layer.sum(axis=1)
+        exp_row_sums = exp_layer.sum(axis=1)
+
+    is_normalized = np.allclose(row_sums, 1)
+    is_log1p_normalized = np.isfinite(exp_row_sums).all() and np.allclose(
+        exp_row_sums, exp_row_sums[0]
+    )
 
-    return np.allclose(row_sums, 1)
+    return is_normalized or is_log1p_normalized
 
 
 def count_obs(adata, pb_adata, obs_cols):

src/differential_expression/create_pseudobulk/test.py

@@ -1,6 +1,8 @@
 import os
 import mudata as mu
 import sys
+import re
+import subprocess
 import pytest
 
 ## VIASH START
@@ -41,6 +43,32 @@ def test_simple_execution(run_component, random_h5mu_path):
     assert adata.shape[0] == 8, "Expected a total of 8 pseudobulk samples in the output"
 
 
+def test_log_normalized_counts(run_component, random_h5mu_path):
+    output_path = random_h5mu_path()
+    with pytest.raises(subprocess.CalledProcessError) as err:
+        run_component(
+            [
+                "--input",
+                input_path,
+                "--output",
+                output_path,
+                "--input_layer",
+                "log_normalized",
+                "--obs_label",
+                "cell_type",
+                "--obs_groups",
+                "treatment",
+                "--output_compression",
+                "gzip",
+            ]
+        )
+
+    assert re.search(
+        r"ValueError: Input layer must contain raw counts.",
+        err.value.stdout.decode("utf-8"),
+    )
+
+
 def test_multiple_factors(run_component, random_h5mu_path):
     output_path = random_h5mu_path()
 

src/differential_expression/deseq2/script.R

@@ -59,17 +59,6 @@ h5mu_to_h5ad <- function(h5mu_path, modality_name) {
   tmp_path
 }
 
-# Check if expression data is normalized (row sums =~ 1)
-is_normalized <- function(layer) {
-  row_sums <- if (is(layer, "sparseMatrix") || is(layer, "dgCMatrix")) {
-    Matrix::rowSums(layer)
-  } else {
-    rowSums(layer)
-  }
-
-  all(abs(row_sums - 1) < 1e-6, na.rm = TRUE)
-}
-
 # Extract design factors from formula
 parse_design_formula <- function(design_formula) {
   if (!grepl("^~\\s*\\w+(\\s*\\+\\s*\\w+)*$", design_formula)) {
@@ -293,10 +282,6 @@ main <- function() {
     mod$X
   }
 
-  if (is_normalized(layer)) {
-    stop("Input layer must contain raw counts.")
-  }
-
   # Prepare analysis components
   cat("Preparing design formula\n")
   design_factors <- parse_design_formula(par$design_formula)

src/differential_expression/deseq2/test.py

@@ -267,7 +267,6 @@ def test_invalid_contrast_column(run_component, tmp_path, pseudobulk_test_data_p
             ]
         )
 
-    # Updated regex to match actual R error format (no square brackets)
     assert re.search(
         r"Missing required columns in metadata: nonexistent_column",
         err.value.stdout.decode("utf-8"),
@@ -296,7 +295,6 @@ def test_invalid_design_column(run_component, tmp_path, pseudobulk_test_data_pat
             ]
         )
 
-    # Updated regex to match actual R error format (no square brackets)
     assert re.search(
         r"Invalid design formula: 'malformed formula'",
         err.value.stdout.decode("utf-8"),