Backport PR #2698 on branch 1.9.x (Fix for highly_variable_genes flavor=seurat modifying layer)

docs/release-notes/1.9.6.md

@@ -5,4 +5,6 @@
 
 - Allow {func}`scanpy.pl.scatter` to accept a {class}`str` palette name {pr}`2571` {smaller}`P Angerer`
 - Make {func}`scanpy.external.tl.palantir` compatible with palantir >=1.3 {pr}`2672` {smaller}`DJ Otto`
-- Fix {func}`scanpy.pl.pca` when when `return_fig=True` and `annotate_var_explained=True` {pr}`2682` {smaller}`J Wagner`
+- Fix {func}`scanpy.pl.pca` when `return_fig=True` and `annotate_var_explained=True` {pr}`2682` {smaller}`J Wagner`
+- Temp fix for {issue}`2680` by skipping `seaborn` version 0.13.0 {pr}`2661` {smaller}`P Angerer`
+- Fix {func}`scanpy.pp.highly_variable_genes` to not modify the used layer when `flavor=seurat` {pr}`2698` {smaller}`E Roellin`

scanpy/preprocessing/_highly_variable_genes.py

@@ -194,9 +194,14 @@ def _highly_variable_genes_single_batch(
     """
     X = adata.layers[layer] if layer is not None else adata.X
     if flavor == 'seurat':
+        X = X.copy()
         if 'log1p' in adata.uns_keys() and adata.uns['log1p'].get('base') is not None:
             X *= np.log(adata.uns['log1p']['base'])
-        X = np.expm1(X)
+        # use out if possible. only possible since we copy X
+        if isinstance(X, np.ndarray):
+            np.expm1(X, out=X)
+        else:
+            X = np.expm1(X)
 
     mean, var = materialize_as_ndarray(_get_mean_var(X))
     # now actually compute the dispersion

scanpy/tests/test_highly_variable_genes.py

@@ -72,6 +72,26 @@ def test_highly_variable_genes_basic():
     assert np.all(np.isin(colnames, hvg_df.columns))
 
 
+@pytest.mark.parametrize('base', [None, 10])
+@pytest.mark.parametrize('flavor', ['seurat', 'cell_ranger'])
+def test_highly_variable_genes_keep_layer(base, flavor):
+    adata = pbmc3k()
+    # cell_ranger flavor can raise error if many 0 genes
+    sc.pp.filter_genes(adata, min_counts=1)
+
+    sc.pp.log1p(adata, base=base)
+    X_orig = adata.X.copy()
+
+    if flavor == 'seurat':
+        sc.pp.highly_variable_genes(adata, n_top_genes=50, flavor=flavor)
+    elif flavor == 'cell_ranger':
+        sc.pp.highly_variable_genes(adata, flavor=flavor)
+    else:
+        assert False
+
+    assert np.allclose(X_orig.A, adata.X.A)
+
+
 def _check_pearson_hvg_columns(output_df, n_top_genes):
     assert pd.api.types.is_float_dtype(output_df['residual_variances'].dtype)