Use Spatialdata IO to Read Visium

pyproject.toml

@@ -44,7 +44,8 @@ dependencies = [
         "tqdm>=4.64.0,<5",
         "shapely>=1.8.0,<2.1",
         "sksparse-minimal>=0.2",
-        "geopandas>=0.14.1,<1"
+        "geopandas>=0.14.1,<1",
+        "spatialdata-io>=0.0.9,<1"
 ]
 
 [project.optional-dependencies]

src/bayestme/data.py

@@ -9,6 +9,7 @@
 import pandas as pd
 import scipy.io as io
 import scipy.sparse.csc
+import spatialdata_io
 from scipy.sparse import csr_matrix
 
 from bayestme import utils
@@ -243,61 +244,18 @@ def read_spaceranger(cls, data_path):
             3) /spatial for position list
         :return: SpatialExpressionDataset
         """
-        raw_count_path = os.path.join(data_path, "raw_feature_bc_matrix/matrix.mtx.gz")
-        filtered_count_path = os.path.join(
-            data_path, "filtered_feature_bc_matrix/matrix.mtx.gz"
-        )
-        features_path = os.path.join(data_path, "raw_feature_bc_matrix/features.tsv.gz")
-        barcodes_path = os.path.join(data_path, "raw_feature_bc_matrix/barcodes.tsv.gz")
-
-        tissue_positions_lists = glob.glob(
-            os.path.join(data_path, "spatial/tissue_positions*.*")
-        )
-
-        positions_path = [fn for fn in tissue_positions_lists if is_csv_tsv(fn)][0]
-
-        if is_tsv(positions_path):
-            positions_df = pd.read_csv(
-                positions_path, sep="\t", header=0, index_col=0, names=None
-            )
-        elif is_csv(positions_path):
-            positions_df = pd.read_csv(
-                positions_path, header=0, index_col=0, names=None
-            )
-        else:
-            raise RuntimeError("No positions list found in spaceranger directory")
-
-        raw_count = np.array(io.mmread(raw_count_path).todense())
-        filtered_count = np.array(io.mmread(filtered_count_path).todense())
-        features = np.array(pd.read_csv(features_path, header=None, sep="\t"))[
-            :, 1
-        ].astype(str)
-        barcodes = pd.read_csv(barcodes_path, header=None, sep="\t")
-        n_spots = raw_count.shape[1]
-        n_genes = raw_count.shape[0]
-        logger.info("detected {} spots, {} genes".format(n_spots, n_genes))
-        positions = positions_df.loc[barcodes[0]][
-            [POSITIONS_X_COLUMN, POSITIONS_Y_COLUMN]
-        ].to_numpy()
-        tissue_mask = positions_df[IN_TISSUE_ATTR].to_numpy().astype(bool)
-        n_spot_in = tissue_mask.sum()
-        logger.info("\t {} spots in tissue sample".format(n_spot_in))
-        all_counts = raw_count.sum()
-        tissue_counts = filtered_count.sum()
-        logger.info(
-            "\t {:.3f}% UMI counts bleeds out".format(
-                (1 - tissue_counts / all_counts) * 100
-            )
-        )
+        sd = spatialdata_io.visium(data_path, dataset_id="visium")
+        ad = sd.table
 
+        tissue_mask = ad.obs.in_tissue.astype(bool).values
         return cls.from_arrays(
-            raw_counts=raw_count.T,
-            positions=positions,
+            raw_counts=ad.X,
+            positions=ad.obs[["array_row", "array_col"]].values,
             tissue_mask=tissue_mask,
-            gene_names=features,
+            gene_names=ad.var_names.values,
             layout=Layout.HEX,
-            edges=utils.get_edges(positions[tissue_mask], Layout.HEX),
-            barcodes=barcodes[0].to_numpy(),
+            edges=utils.get_edges(ad.obsm["spatial"][tissue_mask], Layout.HEX),
+            barcodes=ad.obs_names.values,
         )
 
     @classmethod