MACS3 3.0.5

v3.0.5

on Python PyPI

MACS3 3.0.5

MACS3 3.0.5 improves peak-calling performance, adds a Python API for single-cell fragment analysis, and fixes several peak-calling edge cases. Thank you to the users who reported issues, tested fixes, and contributed code and documentation.

Highlights

• Added PETrackII.return_anndata(regions). It builds an in-memory, sparse barcode-by-region fragment-count matrix as an AnnData object, with cells as rows and genomic regions as columns. It does not automatically write a file. This work includes the initial implementation, tests, and example notebook contributed by @saipenikalapati(#727).
• Optimized NumPy-backed pileup routines used by peak calling and peak-model construction. On our reference callpeak benchmark, 3.0.5 took 27.8% less wall time than 3.0.4—approximately 1.4× faster. Results will vary by dataset and machine.
• Added hmmratac --jump to control fragment-length EM updates.

Fixes

• Preserved decimal region scores in bdgdiff (#739).
• Fixed an IndexError during hmmratac peak refinement (#736) and made successful --cutoff-analysis-only runs exit with status 0 (#731).
• Fixed summit selection in below-cutoff gaps (#741), summit coordinates near peak boundaries (#747), and summit shape filtering (#748).

Compatibility

MACS3 now requires Python 3.12 or later. The cykhash dependency has been removed; pandas and anndata are now runtime dependencies.

Community contributions

Special thanks to @saipenikalapati(#727) for work on the AnnData API and example notebook; @cindykrafft(#739) for the bdgdiff score fix; @kingston487(#736) for the peak-refinement fix; @AbeKline(#731) for the cutoff-analysis exit fix; and @pikammmmm(#728) for improving the bdgcmp documentation.

Full changes since v3.0.4

This revises the draft only; I haven’t changed the tag or published a GitHub Release.