Merge pull request #181 from dpeerlab/dominik/pandas3-cow-fix-issue180

Dominik/pandas3 cow fix issue180

Author: katosh

README.md

@@ -74,6 +74,7 @@ Release Notes
 -------------
 
  ### Version 1.4.5 (future)
+ * Fix: pandas 3 compatibility. Under pandas 3 (mandatory Copy-on-Write) `DataFrame.values` returns a read-only view, so several in-place mutations raised `ValueError: assignment destination is read-only`. `run_palantir` failed in `_differentiation_entropy` (terminal-state identity block, now built with `np.eye(...)`), and `select_branch_cells` failed on its NaN fill of the fate-probability array (now copied before mutation). Compatible with both pandas 2 and 3 ([#180](https://github.com/dpeerlab/Palantir/issues/180)).
  * Use `numpy.random.Generator` in place of the legacy global `numpy.random.RandomState` and require
 `numpy>=1.17`. Note that since the default PRNG is now [PCG64](https://numpy.org/devdocs/reference/random/bit_generators/pcg64.html) instead of the Mersenne Twister, numerical outputs are expected to differ from
 those in previous versions. If exact reproducibility is required, users should pin the relevant prior version.

src/palantir/core.py

@@ -713,8 +713,11 @@ def _differentiation_entropy(
     Q = T[trans_states, :][:, trans_states]
     if len(trans_states) == 0:
         ent = pd.Series(0, index=terminal_states)
-        bp = pd.DataFrame(0, index=terminal_states, columns=terminal_states)
-        bp.values[range(len(terminal_states)), range(len(terminal_states))] = 1
+        bp = pd.DataFrame(
+            np.eye(len(terminal_states)),
+            index=terminal_states,
+            columns=terminal_states,
+        )
         return ent, bp
 
     # Fundamental matrix solver
@@ -761,8 +764,11 @@ def _differentiation_entropy(
 
     # Add terminal states
     ent = pd.concat([ent, pd.Series(0, index=terminal_states)])
-    bp = pd.DataFrame(0, index=terminal_states, columns=terminal_states)
-    bp.values[range(len(terminal_states)), range(len(terminal_states))] = 1
+    bp = pd.DataFrame(
+        np.eye(len(terminal_states)),
+        index=terminal_states,
+        columns=terminal_states,
+    )
     branch_probs = pd.concat([branch_probs, bp.loc[:, branch_probs.columns]])
 
     return ent, branch_probs

src/palantir/presults.py

@@ -528,7 +528,7 @@ def cluster_gene_trends(
         columns=trends.columns,
     )
 
-    gt_ad = AnnData(trends.values, dtype=np.float32)
+    gt_ad = AnnData(trends.values.astype(np.float32))
     sc.pp.neighbors(gt_ad, n_neighbors=n_neighbors, use_rep="X")
 
     # Add required kwargs for leiden with igraph backend to avoid FutureWarning
@@ -604,6 +604,10 @@ def select_branch_cells(
     fate_probs, fate_names = _validate_obsm_key(ad, fate_prob_key, as_df=False)
     pseudotime = ad.obs[pseudo_time_key].values
 
+    # Own the array before the in-place NaN fill: when obsm holds a DataFrame,
+    # _validate_obsm_key returns a .values view that is read-only under pandas 3
+    # (mandatory Copy-on-Write).
+    fate_probs = fate_probs.copy()
     fate_probs[np.isnan(fate_probs)] = 1 / fate_probs.shape[1]
 
     idx = np.argsort(pseudotime)