add squidpy backend

docs/conf.py

@@ -79,6 +79,7 @@
     "pylibraft",
     "dask",
     "cuvs",
+    "spatialdata",
 ]
 default_role = "literal"
 napoleon_google_docstring = False
@@ -126,6 +127,7 @@
     "statsmodels": ("https://www.statsmodels.org/stable/", None),
     "omnipath": ("https://omnipath.readthedocs.io/en/latest/", None),
     "dask": ("https://docs.dask.org/en/stable/", None),
+    "spatialdata": ("https://spatialdata.scverse.org/en/stable/", None),
 }
 
 # List of patterns, relative to source directory, that match files and

pyproject.toml

@@ -71,6 +71,9 @@ dev = [
     "pre-commit",
 ]
 
+[project.entry-points."squidpy.backends"]
+rapids_singlecell = "rapids_singlecell.squidpy_backend:RscSquidpyBackend"
+
 [project.urls]
 Documentation = "https://rapids-singlecell.readthedocs.io"
 Source = "https://github.com/scverse/rapids_singlecell"

src/rapids_singlecell/_compat.py

@@ -7,6 +7,11 @@
 from scipy.sparse import csc_matrix as csc_matrix_cpu
 from scipy.sparse import csr_matrix as csr_matrix_cpu
 
+try:
+    from spatialdata import SpatialData
+except ImportError:
+    SpatialData = None
+
 
 def _meta_dense(dtype):
     return cp.zeros([0], dtype=dtype)

src/rapids_singlecell/squidpy_backend.py

@@ -0,0 +1,30 @@
+"""Squidpy backend adapter for rapids_singlecell.
+
+The dispatch decorator introspects the real RSC function signatures
+(lazily imported on first access), so no need to duplicate them here.
+"""
+
+from __future__ import annotations
+
+import importlib
+
+
+class RscSquidpyBackend:
+    """Backend adapter exposing rapids_singlecell GPU implementations to squidpy."""
+
+    name = "rapids_singlecell"
+    aliases = ["rapids-singlecell", "rsc", "cuda", "gpu"]
+
+    # squidpy function name -> module that implements it
+    _functions = {
+        "spatial_autocorr": "rapids_singlecell.squidpy_gpu",
+        "co_occurrence": "rapids_singlecell.squidpy_gpu",
+        "ligrec": "rapids_singlecell.squidpy_gpu",
+    }
+
+    def __getattr__(self, name: str):
+        if name in self._functions:
+            func = getattr(importlib.import_module(self._functions[name]), name)
+            setattr(self, name, func)  # cache on instance
+            return func
+        raise AttributeError(f"{type(self).__name__!r} has no attribute {name!r}")

src/rapids_singlecell/squidpy_gpu/_autocorr.py

@@ -12,6 +12,7 @@
 from scipy import sparse
 from statsmodels.stats.multitest import multipletests
 
+from rapids_singlecell._compat import SpatialData
 from rapids_singlecell.preprocessing._utils import _sparse_to_dense
 
 from ._gearysc import _gearys_C_cupy
@@ -49,7 +50,7 @@ def _to_cupy(vals, *, use_sparse: bool, dtype):
 
 
 def spatial_autocorr(
-    adata: AnnData,
+    adata: AnnData | SpatialData,
     *,
     connectivity_key: str = "spatial_connectivities",
     genes: str | Sequence[str] | None = None,
@@ -118,6 +119,8 @@ def spatial_autocorr(
             DataFrame containing the autocorrelation scores, p-values, and corrected p-values for each gene. \
             If `copy` is False, the results are stored in `adata.uns` and None is returned.
     """
+    if SpatialData is not None and isinstance(adata, SpatialData):
+        adata = adata.table
     if genes is None:
         if "highly_variable" in adata.var:
             genes = adata[:, adata.var["highly_variable"]].var_names.values

src/rapids_singlecell/squidpy_gpu/_co_oc.py

@@ -6,6 +6,7 @@
 import numpy as np
 from cuml.metrics import pairwise_distances
 
+from rapids_singlecell._compat import SpatialData
 from rapids_singlecell._cuda import _cooc_cuda as _co
 from rapids_singlecell._utils import (
     _calculate_blocks_per_pair,
@@ -21,7 +22,7 @@
 
 
 def co_occurrence(
-    adata: AnnData,
+    adata: AnnData | SpatialData,
     cluster_key: str,
     *,
     spatial_key: str = "spatial",
@@ -65,6 +66,8 @@ def co_occurrence(
           computed at ``interval``.
     """
 
+    if SpatialData is not None and isinstance(adata, SpatialData):
+        adata = adata.table
     _assert_categorical_obs(adata, key=cluster_key)
     _assert_spatial_basis(adata, key=spatial_key)
     spatial = cp.array(adata.obsm[spatial_key]).astype(np.float32)

src/rapids_singlecell/squidpy_gpu/_ligrec.py

@@ -14,6 +14,8 @@
 from cupyx.scipy.sparse import issparse as cpissparse
 from scipy.sparse import csc_matrix, issparse
 
+from rapids_singlecell._compat import SpatialData
+
 from ._utils import _assert_categorical_obs, _create_sparse_df
 
 SOURCE = "source"
@@ -118,7 +120,7 @@ def _check_tuple_needles(needles, haystack, *, msg: str, reraise: bool = True):
 
 
 def ligrec(
-    adata: AnnData,
+    adata: AnnData | SpatialData,
     cluster_key: str,
     *,
     clusters: list | None = None,
@@ -233,6 +235,8 @@ def ligrec(
     interacting components was 0 or it didn't pass the threshold percentage of \
     cells being expressed within a given cluster.
     """
+    if SpatialData is not None and isinstance(adata, SpatialData):
+        adata = adata.table
     # Get and Check interactions
     if interactions is None:
         interactions = _get_interactions(

tests/test_backend_conformance.py

@@ -0,0 +1,11 @@
+"""Run squidpy's backend conformance suite against the RSC backend."""
+
+from __future__ import annotations
+
+from squidpy.testing.backend_conformance import validate_backend
+
+
+def test_conformance():
+    results = validate_backend("rapids_singlecell")
+    for name, status in results.items():
+        assert status == "PASSED", f"{name}: {status}"