fix: replace underscores in factor names or factor levels with hyphens

pydeseq2/dds.py

@@ -213,6 +213,33 @@ def __init__(
             raise ValueError("NaNs are not allowed in the design factors.")
         self.obs[self.design_factors] = self.obs[self.design_factors].astype(str)
 
+        # Check that design factors don't contain underscores. If so, convert them to
+        # hyphens.
+        if np.any(["_" in factor for factor in self.design_factors]):
+            warnings.warn(
+                """Same factor names in the design contain underscores ('_'). They will
+                be converted to hyphens ('-').""",
+                UserWarning,
+                stacklevel=2,
+            )
+
+            new_factors = [factor.replace("_", "-") for factor in self.design_factors]
+
+            self.obs.rename(
+                columns={
+                    old_factor: new_factor
+                    for (old_factor, new_factor) in zip(self.design_factors, new_factors)
+                },
+                inplace=True,
+            )
+
+            self.design_factors = new_factors
+
+        # If ref_level has underscores, covert them to hyphens
+        # Don't raise a warning: it will be raised by build_design_matrix()
+        if ref_level is not None:
+            ref_level = [name.replace("_", "-") for name in ref_level]
+
         # Build the design matrix
         # Stored in the obsm attribute of the dataset
         self.obsm["design_matrix"] = build_design_matrix(

pydeseq2/utils.py

@@ -1,4 +1,5 @@
 import multiprocessing
+import warnings
 from math import floor
 from pathlib import Path
 from typing import List
@@ -192,6 +193,21 @@ def build_design_matrix(
                 f"takes the single value '{np.unique(metadata[factor])}'."
             )
 
+    # Check that level factors in the design don't contain underscores. If so, convert
+    # them to hyphens
+    warning_issued = False
+    for factor in design_factors:
+        if np.any(["_" in value for value in metadata[factor]]):
+            if not warning_issued:
+                warnings.warn(
+                    """Some factor levels in the design contain underscores ('_').
+                    They will be converted to hyphens ('-').""",
+                    UserWarning,
+                    stacklevel=2,
+                )
+                warning_issued = True
+            metadata[factor] = metadata[factor].apply(lambda x: x.replace("_", "-"))
+
     design_matrix = pd.get_dummies(metadata[design_factors], drop_first=not expanded)
 
     if ref_level is not None:

tests/test_edge_cases.py

@@ -136,12 +136,14 @@ def test_underscores_in_factors():
         inplace=True,
     )
 
-    # Run the pipeline. This should not cause bugs.
-    dds = DeseqDataSet(
-        counts=counts_df,
-        metadata=metadata,
-        design_factors="some_variable_with_underscores",
-    )
+    # Run the pipeline. This should raise a warning, but not cause bugs.
+    with pytest.warns(UserWarning):
+        dds = DeseqDataSet(
+            counts=counts_df,
+            metadata=metadata,
+            design_factors="some_variable_with_underscores",
+            ref_level=["some_variable_with_underscores", "level_with_underscores"],
+        )
     dds.deseq2()
 
     stat_res = DeseqStats(dds)