DO NOT MERGE - set a default large component threshold

[ptajvar]

Removing large components by default in the graph step. This should eventually be an option in the pipeline and not done by default.

Fixes: PNA-3188

Type of change

TEMP

---

Note

High Risk
Default graph behavior now hard-filters components by size (including dropping clusters >500k UMIs), which can materially change cell counts and downstream PXL output without explicit user configuration; marked TEMP in the PR description.

Overview
The graph CLI now defaults --component-size-max-threshold to 500,000 and --component-size-min-threshold to MIN_PNA_COMPONENT_SIZE, so runs use hard min/max bounds instead of dynamic size filtering unless callers change those flags.

Early hive-partitioned edgelist writing was extended from minimum-only pruning to min and max UMI bounds: write_hive_partitioned_edgelist_without_out_of_size_bound_components drops components below the min or above the max (same UMI score as hybrid detection). find_components passes the upper bound from component_size_threshold into that step and logs the applied min/max; the default upper bound type uses np.iinfo(np.uint64).max instead of 2**32 - 1.

^{Reviewed by Cursor Bugbot for commit 9375ec8. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes using high effort and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit dde93a6. Configure here.}

[cursor]

Max prune empty graph mishandled

Medium Severity

With the new default max component UMI bound, a graph where every hybrid component is above that limit can pass the pre-hive size check, then lose all edges in write_hive_partitioned_edgelist_without_out_of_size_bound_components. The pipeline keeps going instead of failing clearly: with multiplet recovery enabled, calculate_post_recovery_component_statistics can hit an empty component table when computing fraction_nodes_in_largest_component_post_recovery; otherwise the run fails later with a message that blames components being too small.

Additional Locations (1)

• src/pixelator/pna/graph/community_detection.py#L145-L149

 

^{Reviewed by Cursor Bugbot for commit dde93a6. Configure here.}