Backport PR #4112 on branch 1.12.x (fix: use get_igraph_from_adjacency for graph construction)

docs/release-notes/4112.fix.md

@@ -0,0 +1,4 @@
+Make {func}`scanpy.metrics.modularity` produce the exact same results as `igraph` itself,
+see `https://github.com/scverse/scanpy/blob/e4e43830ea3d5af0acd99648612612551fa7b83e/tests/test_metrics.py#L340 <here>`__.
+In Scanpy 2.0, all graphs will be calculated slightly differently, using {meth}`igraph.Graph.Weighted_Adjacency`.
+{smaller}`P Angerer`

src/scanpy/_utils/__init__.py

@@ -74,6 +74,7 @@
     "compute_association_matrix_of_groups",
     "descend_classes_and_funcs",
     "ensure_igraph",
+    "get_igraph_from_adjacency",
     "get_literal_vals",
     "indent",
     "is_backed_type",
@@ -287,7 +288,7 @@ def check_use_raw(
 # --------------------------------------------------------------------------------
 
 
-def get_igraph_from_adjacency(adjacency: CSBase, *, directed: bool = False) -> Graph:
+def get_igraph_from_adjacency(adjacency: CSBase, *, directed: bool) -> Graph:
     """Get igraph graph from adjacency matrix."""
     import igraph as ig
 

src/scanpy/metrics/_metrics.py

@@ -10,7 +10,7 @@
 from natsort import natsorted
 from pandas.api.types import CategoricalDtype
 
-from .._utils import NeighborsView
+from .._utils import NeighborsView, get_igraph_from_adjacency
 
 if TYPE_CHECKING:
     from collections.abc import Sequence
@@ -203,15 +203,14 @@ def modularity_array(
     connectivities: AnyArrayLike | SpBase, /, *, labels: AnyArrayLike, is_directed: bool
 ) -> float:
     try:
-        import igraph as ig
+        import igraph  # noqa: F401
     except ImportError as e:  # pragma: no cover
         e.add_note(
             "`igraph` is required for computing modularity. "
             "Please install `igraph` and try again."
         )
         raise
-    igraph_mode: str = ig.ADJ_DIRECTED if is_directed else ig.ADJ_UNDIRECTED
-    graph: ig.Graph = ig.Graph.Weighted_Adjacency(connectivities, mode=igraph_mode)
+    graph = get_igraph_from_adjacency(connectivities, directed=is_directed)
     return graph.modularity(_codes(labels), "weight")
 
 

src/scanpy/neighbors/__init__.py

@@ -571,7 +571,7 @@ def distances_dpt(self) -> OnFlySymMatrix:
 
     def to_igraph(self) -> Graph:
         """Generate igraph from connectiviies."""
-        return _utils.get_igraph_from_adjacency(self.connectivities)
+        return _utils.get_igraph_from_adjacency(self.connectivities, directed=False)
 
     @_doc_params(n_pcs=doc_n_pcs, use_rep=doc_use_rep)
     def compute_neighbors(

src/scanpy/plotting/_tools/paga.py

@@ -259,7 +259,7 @@ def _compute_pos(  # noqa: PLR0912
     else:
         # igraph layouts
         random.seed(random_state.bytes(8))
-        g = _sc_utils.get_igraph_from_adjacency(adjacency_solid)
+        g = _sc_utils.get_igraph_from_adjacency(adjacency_solid, directed=False)
         if "rt" in layout:
             g_tree = _sc_utils.get_igraph_from_adjacency(adj_tree)
             pos_list = g_tree.layout(

src/scanpy/tools/_draw_graph.py

@@ -151,7 +151,7 @@ def draw_graph(  # noqa: PLR0913
         # igraph doesn't use numpy seed
         random.seed(random_state)
 
-        g = _utils.get_igraph_from_adjacency(adjacency)
+        g = _utils.get_igraph_from_adjacency(adjacency, directed=False)
         if layout in {"fr", "drl", "kk", "grid_fr"}:
             ig_layout = g.layout(layout, seed=init_coords.tolist(), **kwds)
         elif "rt" in layout:

src/scanpy/tools/_paga.py

@@ -214,7 +214,7 @@ def _compute_connectivities_v1_0(self):
 
         ones = self._neighbors.connectivities.copy()
         ones.data = np.ones(len(ones.data))
-        g = _utils.get_igraph_from_adjacency(ones)
+        g = _utils.get_igraph_from_adjacency(ones, directed=False)
         vc = igraph.VertexClustering(
             g, membership=self._adata.obs[self._groups_key].cat.codes.values
         )

src/scanpy/tools/_utils.py

@@ -54,12 +54,13 @@ def _get_pca_or_small_x(adata: AnnData, n_pcs: int | None) -> np.ndarray | CSRBa
         logg.info("    using data matrix X directly")
         return adata.X
 
-    if "X_pca" in adata.obsm:
-        if n_pcs is not None and n_pcs > adata.obsm["X_pca"].shape[1]:
-            msg = "`X_pca` does not have enough PCs. Rerun `sc.pp.pca` with adjusted `n_comps`."
+    pca_key = next((k for k in ("pca", "X_pca") if k in adata.obsm), None)
+    if pca_key is not None:
+        if n_pcs is not None and n_pcs > adata.obsm[pca_key].shape[1]:
+            msg = f"`{pca_key}` does not have enough PCs. Rerun `sc.pp.pca` with adjusted `n_comps`."
             raise ValueError(msg)
-        x = adata.obsm["X_pca"][:, :n_pcs]
-        logg.info(f"    using 'X_pca' with n_pcs = {x.shape[1]}")
+        x = adata.obsm[pca_key][:, :n_pcs]
+        logg.info(f"    using {pca_key!r} with n_pcs = {x.shape[1]}")
         return x
 
     from ..preprocessing import pca

tests/test_metrics.py

@@ -334,7 +334,7 @@ def test_modularity_adata(
             assert 0 <= s <= 1
     for (n0, s0), (n1, s1) in combinations(scores.items(), 2):
         with subtests.test("equality", l=n0, r=n1):
-            assert pytest.approx(s0, rel=1e-6) == s1
+            assert s0 == s1
     with subtests.test("update"):
         assert adata.uns["leiden"]["modularity"] is scores["update"]