SWE-bench-Science / tasks /task_018 /instruction.md
Kevin589981's picture
publish SWE-bench Science release
91895bb verified
|
Raw
History Blame Contribute Delete
1.73 kB

Repair a scientifically inconsistent HMMRATAC workflow

The supplied directory contains a pinned snapshot of MACS3, including the real HMMRATAC implementation used to infer accessible chromatin from ATAC-seq fragments. The offline reproduction exposes an unexpected behavior: adding a genomic exclusion mask can change model-input statistics and derived behavior on chromosomes that are not covered by the mask.

Inspect the HMMRATAC documentation in source/ and the complete fragment-to-model call graph, and repair the implementation so the exclusion mask has the intended scientific meaning throughout the workflow. The repair must work for ordinary paired-end fragment data and count-bearing single-cell fragment data, preserve half-open interval semantics, and keep count and barcode bookkeeping consistent for count-bearing fragment data.

Duplicate retention and duplicate-related CLI naming are intentionally outside the scope of this task. Preserve the existing duplicate policy and do not add, remove, rename, or reinterpret duplicate flags. Repeated coordinates in the fixtures are there to verify that blacklist filtering preserves the existing record multiplicity and count/barcode associations.

This is a capability repair, not a fixture-specific patch. Do not hard-code the public chromosomes, coordinates, lengths, counts, or expected outputs. Preserve the existing public workflow and keep all generated files under outputs/.

Run the public reproduction with:

python reproduce.py

Treat the supplied case as one observation of the workflow defect. The repair must preserve the documented scientific and record-association semantics for other valid inputs supported by the same production paths.