fix: resolve kaleido/plotly incompatibility and scipy sparse OOM

- Pin kaleido <1.0.0 since kaleido 1.x requires plotly >=6.1.1
  while the project uses plotly 5.x
- Use two-step sparse matrix indexing in spatial_ldsc to avoid
  scipy's _get_arrayXarray broadcasting a 51.6 GiB dense intermediate

Author: Ganten-Hornby

pyproject.toml

@@ -33,7 +33,7 @@ dependencies = [
     "pyranges >=0.0.120, <1.0.0",
     "pyfiglet >=0.8",
     "plotly >=5.0.0, <6.0.0",
-    "kaleido >=0.2.0, <2.0.0",
+    "kaleido >=0.2.0, <1.0.0",
     "jinja2 >=3.0.0, <4.0.0",
     "scanpy >=1.8.0, <2.0.0",
     "zarr >=2.0.0, <3.0.0",

src/gsMap/spatial_ldsc_multiple_sumstats.py

@@ -177,9 +177,11 @@ def __init__(self, config: SpatialLDSCConfig, common_snp_among_all_sumstats_pos)
         snp_gene_weight_adata = ad.read_h5ad(config.snp_gene_weight_adata_path)
         common_genes = mk_score_genes.intersection(snp_gene_weight_adata.var.index)
         # common_snps = snp_gene_weight_adata.obs.index
-        self.snp_gene_weight_matrix = snp_gene_weight_adata[
-            common_snp_among_all_sumstats_pos, common_genes.to_list()
-        ].X
+        # Two-step indexing to avoid scipy sparse _get_arrayXarray broadcasting OOM
+        col_idx = snp_gene_weight_adata.var.index.get_indexer(common_genes)
+        self.snp_gene_weight_matrix = snp_gene_weight_adata.X[common_snp_among_all_sumstats_pos][
+            :, col_idx
+        ]
         self.mk_score_common = mk_score.loc[common_genes]
         self.chunk_starts = list(
             range(0, self.mk_score_common.shape[1], self.config.spots_per_chunk_quick_mode)