`spatial_neighbors` breaks when `transform='spectral'`

[tjbencomo]

Description

When specifying transform='spectral' in sq.gr.spatial_neighbors, I encounter a ValueError: inconsistent shapes error

Setting transform='cosine' works fine and does not produce an error

This error seems to be reproducible in multiple Xenium datasets

Not sure if the error is due to an incorrect combination of parameter settings or squidpy bug

Minimal reproducible example

Using the Xenium V1 Lung Cancer FFPE dataset used in the Squidpy Xenium tutorial

import os
import matplotlib.pyplot as plt
import scanpy as sc
import seaborn as sns
import spatialdata as sd
import squidpy as sq
from spatialdata_io import xenium

baseDir = "/hpc/temp/setty_m/tbencomo/Xenium_V1_humanLung_Cancer_FFPE"
zarr_path = os.path.join(baseDir, "Xenium.zarr")

sdata = sd.read_zarr(zarr_path)
adata = sdata.tables["table"]

adata.layers["counts"] = adata.X.copy()
sc.pp.normalize_total(adata, inplace=True)
sc.pp.log1p(adata)
sc.pp.pca(adata)
sc.pp.neighbors(adata)
sc.tl.umap(adata)
sc.tl.leiden(adata)

# This works - no issues
sq.gr.spatial_neighbors(adata, coord_type="generic", delaunay=True)

# Also works no issues
sq.gr.spatial_neighbors(
    adata,
    coord_type="generic",
    delaunay=True,
    transform="cosine",
    key_added="spatial_cosine",
)

# This throws an error
sq.gr.spatial_neighbors(
    adata,
    coord_type="generic",
    delaunay=True,
    transform="spectral",
    key_added="spatial_spectral",
)

# Also throws an error
sq.gr.spatial_neighbors(
    adata,
    coord_type="generic",
    transform="spectral",
    key_added="spatial_spectral",
)

# Also throws an error
sq.gr.spatial_neighbors(
    adata,
    delaunay=True,
    transform="spectral",
    key_added="spatial_spectral",
)

# Also errors
sq.gr.spatial_neighbors(
    adata,
    transform="spectral",
    key_added="spatial_spectral",
)

Traceback

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Cell In[24], line 1
----> 1 sq.gr.spatial_neighbors(
      2     adata,
      3     coord_type="generic",
      4     delaunay=True,
      5     transform="spectral",
      6     key_added="spatial_spectral",
      7 )

File [~/micromamba/envs/spatial_v1/lib/python3.12/site-packages/squidpy/gr/_build.py:244](http://gizmok22.fhcrc.org:56252/lab/tree/mcc-spatial/notebooks/micromamba/envs/spatial_v1/lib/python3.12/site-packages/squidpy/gr/_build.py#line=243), in spatial_neighbors(adata, spatial_key, elements_to_coordinate_systems, table_key, library_key, coord_type, n_neighs, radius, delaunay, n_rings, percentile, transform, set_diag, key_added, copy)
    242     Dst = block_diag([m[1] for m in mats], format="csr")[ixs, :][:, ixs]
    243 else:
--> 244     Adj, Dst = _build_fun(adata)
    246 neighs_key = Key.uns.spatial_neighs(key_added)
    247 conns_key = Key.obsp.spatial_conn(key_added)

File [~/micromamba/envs/spatial_v1/lib/python3.12/site-packages/squidpy/gr/_build.py:323](http://gizmok22.fhcrc.org:56252/lab/tree/mcc-spatial/notebooks/micromamba/envs/spatial_v1/lib/python3.12/site-packages/squidpy/gr/_build.py#line=322), in _spatial_neighbor(adata, spatial_key, coord_type, n_neighs, radius, delaunay, n_rings, transform, set_diag, percentile)
    321 # check transform
    322 if transform == Transform.SPECTRAL:
--> 323     Adj = _transform_a_spectral(Adj)
    324 elif transform == Transform.COSINE:
    325     Adj = _transform_a_cosine(Adj)

File [~/micromamba/envs/spatial_v1/lib/python3.12/site-packages/squidpy/gr/_build.py:460](http://gizmok22.fhcrc.org:56252/lab/tree/mcc-spatial/notebooks/micromamba/envs/spatial_v1/lib/python3.12/site-packages/squidpy/gr/_build.py#line=459), in _transform_a_spectral(a)
    457     return a
    459 degrees = np.squeeze(np.array(np.sqrt(1.0 [/](http://gizmok22.fhcrc.org:56252/) a.sum(axis=0))))
--> 460 a = a.multiply(outer(a.indices, a.indptr, degrees))
    461 a.eliminate_zeros()
    463 return a

File [~/micromamba/envs/spatial_v1/lib/python3.12/site-packages/scipy/sparse/_compressed.py:476](http://gizmok22.fhcrc.org:56252/lab/tree/mcc-spatial/notebooks/micromamba/envs/spatial_v1/lib/python3.12/site-packages/scipy/sparse/_compressed.py#line=475), in _cs_matrix.multiply(self, other)
    474     data = np.multiply(ret.data, other2d[ret.row].ravel())
    475 else:
--> 476     raise ValueError("inconsistent shapes")
    477 ret.data = data.view(np.ndarray).ravel()
    478 return ret

ValueError: inconsistent shapes

Version

1.6.5

...

[selmanozleyen]

hi @tjbencomo , thanks for showing this. I noticed that the spectral transformation wasn't tested at and didn't see it working in any notebooks yet. If it was working before I didn't see it. It's crazy that nobody else complained about this.

here is my fix lmk if it works for you as expected: #1028

but maybe I am missing something because this seems like a very obvious error for me