Merge pull request #290 from lilab-bcb/fix-sig-background

Enforce dense or csr_matrix type for calculate_z_score function

pegasus/tools/__init__.py

@@ -3,6 +3,7 @@
     update_rep,
     X_from_rep,
     W_from_rep,
+    to_csr_or_dense,
     slicing,
     calc_mean,
     calc_mean_and_var,

pegasus/tools/signature_score.py

@@ -7,7 +7,7 @@
 
 import anndata
 from pegasusio import UnimodalData, MultimodalData, timer
-from pegasus.tools import calc_mean, calc_sig_background, predefined_signatures, load_signatures_from_file
+from pegasus.tools import to_csr_or_dense, calc_mean, calc_sig_background, predefined_signatures, load_signatures_from_file
 
 import logging
 logger = logging.getLogger(__name__)
@@ -16,7 +16,7 @@
 
 def _check_and_calc_sig_background(data: UnimodalData, n_bins: int) -> bool:
     if "mean" not in data.var:
-        data.var["mean"] = calc_mean(data.X, axis = 0)
+        data.var["mean"] = calc_mean(to_csr_or_dense(data.X),axis = 0)
 
     if data.uns.get("sig_n_bins", 0) != n_bins:
         mean_vec = data.var["mean"].values
@@ -35,7 +35,7 @@ def _check_and_calc_sig_background(data: UnimodalData, n_bins: int) -> bool:
         data.var["bins"] = bins
 
         # calculate background expectations
-        data.obsm["sig_bkg_mean"], data.obsm["sig_bkg_std"] = calc_sig_background(data.X, bins, mean_vec)
+        data.obsm["sig_bkg_mean"], data.obsm["sig_bkg_std"] = calc_sig_background(to_csr_or_dense(data.X), bins, mean_vec)
 
     return True
 
@@ -190,7 +190,7 @@ def calc_signature_score(
                 del signatures["mito_noncoding"]
             elif sig_string.startswith("ribosomal_genes"):
                 del signatures["ribo_like"]
-            
+
             _calc_sig_scores(data, signatures, show_omitted_genes=show_omitted_genes, skip_threshold=skip_threshold)
 
             if sig_string.startswith("cell_cycle"):

pegasus/tools/utils.py

@@ -1,7 +1,7 @@
 import numpy as np
 import pandas as pd
 from pandas.api.types import is_categorical_dtype
-from scipy.sparse import issparse, csr_matrix
+from scipy.sparse import issparse, csr_matrix, csc_matrix
 from typing import Union, List, Tuple, Dict
 from anndata import AnnData
 from pegasusio import UnimodalData, MultimodalData
@@ -61,6 +61,10 @@ def W_from_rep(
     return data.obsp[rep_key]
 
 
+def to_csr_or_dense(X: Union[csr_matrix, csc_matrix, np.ndarray]) -> Union[csr_matrix, np.ndarray]:
+    return X if not issparse(X) or isinstance(X, csr_matrix) else X.tocsr()
+
+
 # slicing is not designed to work at extracting one element, convert to dense matrix
 def slicing(X: Union[csr_matrix, np.ndarray], row: Union[List[bool], List[int], int] = slice(None), col: Union[List[bool], List[int], int] = slice(None), copy: bool = False, squeeze: bool = True) -> np.ndarray:
     result = X[row, col]