[stephenwilliams22]

Adds a native reader for 10x Genomics Atera datasets: cell-by-gene expression table, cell/nucleus segmentation labels and boundary polygons, per-transcript locations, and morphology images.

• Table (cell_feature_matrix.zarr.zip/csc_cell_feature_matrix.zarr.zip) is read directly with anndata's own zarr IO (anndata.io.read_elem/ anndata.io.sparse_dataset), with X lazily backed by dask.
• Cell/nucleus labels are lazily backed by dask; boundary polygons are read via shapely.from_ragged_array, with an optional tiled/pyramided partial-read path (read_cell_boundaries) for large bundles.
• Transcripts are read lazily per spatial tile via dask-delayed.
• Morphology images are read via tifffile's native OME-TIFF tile grid, avoiding materializing full-resolution planes.
• An example notebook with a finalized tiny atera output bundle is provided for testing. This can be found in the atera_example.ipynb for testing.

Enables massive datasets on a basic laptop (2M cells, 11 billion transcripts). Basic analysis workflow (loading, plotting, HVG, PCA, normalization, clustering, etc) max memory usage was 7GB on a new mac in ~15min.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[codecov-commenter]

Codecov Report

❌ Patch coverage is 62.54296% with 218 lines in your changes missing coverage. Please review.
✅ Project coverage is 66.00%. Comparing base (9548686) to head (488acf7).

Files with missing lines	Patch %	Lines
src/spatialdata_io/readers/_atera_common.py	45.92%	166 Missing ⚠️
src/spatialdata_io/readers/atera.py	71.11%	52 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #426      +/-   ##
==========================================
- Coverage   66.60%   66.00%   -0.61%     
==========================================
  Files          26       29       +3     
  Lines        3345     3927     +582     
==========================================
+ Hits         2228     2592     +364     
- Misses       1117     1335     +218     

Files with missing lines	Coverage Δ
src/spatialdata_io/__init__.py	100.00% <ø> (ø)
src/spatialdata_io/_constants/_constants.py	100.00% <100.00%> (ø)
src/spatialdata_io/converters/atera_to_parquet.py	100.00% <100.00%> (ø)
src/spatialdata_io/readers/atera.py	71.11% <71.11%> (ø)
src/spatialdata_io/readers/_atera_common.py	45.92% <45.92%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Zethson]

@stephenwilliams22 could you please always ensure that the pre-commit checks are green? Ideally a PR should always have green CI before anyone has a look.

Thanks!

[stephenwilliams22]

@stephenwilliams22 could you please always ensure that the pre-commit checks are green? Ideally a PR should always have green CI before anyone has a look.

Thanks!

@Zethson sorry about that. we should be good to go now.

[xinaesthete]

@stephenwilliams22 It seems you write attrs such that the table annotates cell_labels, but the raster values there don't correspond to the cell_ids in the table. You could probably associate with cell_boundaries shapes instead, or have another obs['label-ids'] and use that.

I'm going from what Claude has said more than actually having read through the code in detail.

Ideally you'd be able to record an association between more than one somewhat equivalent spatial element (so a given cell could be associated with nucleus & boundary shapes/labels), I've been meaning to find if there's an open issue or discussion on that.

[xinaesthete]

Also sdata.write() fails on the reader’s output, at the table:

ValueError: DataFrame.index.name ('barcode') is also used by a column whose values are different.

var fails the same way with feature_id. The values aren’t actually different: every barcode and feature ID matches its column. Under pandas 3 the index is the string dtype (str) while the column is object, and anndata checks the pair with pd.Series.equals, which also compares dtypes:

s = pd.Series(obs.index, index=obs.index)
s.equals(obs["barcode"])                          # False
s.equals(obs["barcode"].astype(obs.index.dtype))  # True

Building the index and the column from the same array (so their dtypes match), or not naming the index after an existing column, fixes it. Seen with the Atera v1 WTA kidney dataset; spatialdata 0.8.0, anndata 0.13.4, pandas 3.1.0.

[stephenwilliams22]

@stephenwilliams22 It seems you write attrs such that the table annotates cell_labels, but the raster values there don't correspond to the cell_ids in the table. You could probably associate with cell_boundaries shapes instead, or have another obs['label-ids'] and use that.

I'm going from what Claude has said more than actually having read through the code in detail.

Ideally you'd be able to record an association between more than one somewhat equivalent spatial element (so a given cell could be associated with nucleus & boundary shapes/labels), I've been meaning to find if there's an open issue or discussion on that.

Thanks @xinaesthete I put in a fix

[stephenwilliams22]

Also sdata.write() fails on the reader’s output, at the table:

ValueError: DataFrame.index.name ('barcode') is also used by a column whose values are different.

var fails the same way with feature_id. The values aren’t actually different: every barcode and feature ID matches its column. Under pandas 3 the index is the string dtype (str) while the column is object, and anndata checks the pair with pd.Series.equals, which also compares dtypes:

s = pd.Series(obs.index, index=obs.index)
s.equals(obs["barcode"])                          # False
s.equals(obs["barcode"].astype(obs.index.dtype))  # True

Building the index and the column from the same array (so their dtypes match), or not naming the index after an existing column, fixes it. Seen with the Atera v1 WTA kidney dataset; spatialdata 0.8.0, anndata 0.13.4, pandas 3.1.0.

Fixed