Fix Ingest truncating to settings.N_PCS even when pp.neighbors used more PCs

[SID-6921]

Ingest falls back to a hardcoded settings.N_PCS (50) whenever the reference adata's neighbors params recorded neither use_rep nor n_pcs — but that's exactly the state pp.neighbors leaves when called with its own defaults, and its default (n_pcs=None) uses the full stored PCA representation unsliced (_get_pca_or_small_x), not 50 components. Whenever more than 50 PCs were computed (e.g. sc.pp.pca(adata, n_comps=100) then sc.pp.neighbors(adata)), Ingest silently builds its KNN index and transforms new data in a different, truncated representation than the one that actually produced the reference neighbor graph — no error, just a quietly wrong embedding/label mapping for sc.tl.ingest.

Found via sibling comparison between Ingest._init_neighbors's three branches and _get_pca_or_small_x (the function that actually computes pp.neighbors' representation) — the first two branches already match the stored params correctly; only the fallback diverged.

Repro

import anndata as ad, numpy as np, scanpy as sc
X = np.random.default_rng(0).normal(size=(300, 200)).astype("float32")
adata_ref = ad.AnnData(X)
sc.pp.pca(adata_ref, n_comps=100)
sc.pp.neighbors(adata_ref)  # default n_pcs=None -> uses all 100 PCs
ing = sc.tl.Ingest(adata_ref)
ing._init_neighbors(adata_ref, None)
print(ing._n_pcs)  # 50 before this fix, should be 100

AI assistance

• Tool: Claude Code
• Role: found the inconsistency by comparing Ingest's three _init_neighbors branches against _get_pca_or_small_x's actual default behavior, then wrote the fix and regression test
• Verified locally against a synthetic reproduction (the repo's pbmc68k_reduced test fixture hits an unrelated zarr-version incompatibility in this environment, so the new test couldn't be run through the full suite here — logic was verified directly against the same sc.pp.pca/sc.pp.neighbors/sc.tl.Ingest calls the test uses)

Test results

Added test_representation_more_than_n_pcs_default to tests/test_ingest.py, covering n_comps = settings.N_PCS + 20. Manually verified the underlying fix across n_comps of 30, 50, and 100 against a synthetic AnnData (below, below+at, and above the settings.N_PCS threshold) — all three now correctly match the PCs pp.neighbors actually used, whereas 100 previously and incorrectly truncated to 50.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.78%. Comparing base (7ad567d) to head (52bf151).
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4401   +/-   ##
=======================================
  Coverage   83.78%   83.78%           
=======================================
  Files         134      134           
  Lines       12691    12691           
=======================================
  Hits        10633    10633           
  Misses       2058     2058           

Flag	Coverage Δ
hatch-test.low-vers	80.49% <100.00%> (ø)
hatch-test.pre	83.58% <100.00%> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
src/scanpy/tools/_ingest.py	85.47% <100.00%> (ø)