MACS3 3.1.0 release notes

MACS3 3.1.0 brings a substantial speedup to callpeak, adds controls for CPU use, and improves handling of several input and failure cases.

Faster peak calling

Jacob Schreiber's auto-optimize contribution in PR #756 measured 41–64× faster callpeak runs in an 8-core benchmark. The reported speedups are medians of three runs for each workflow:

These results are workload-specific; short jobs may see little change, and large inputs may use more peak memory.

CPU controls

• Use --threads N on callpeak, predictd, pileup, filterdup, randsample, refinepeak, hmmratac, or callvar to set an invocation CPU budget.
• Set MACS3_NUM_THREADS to provide a default budget when --threads is omitted.
• Pass an execution policy to supported Python APIs to control CPU use from code.
• Use --process-backend {auto,serial} with callpeak or callvar to choose whether process workers may be used.

Fixes

• Automatic duplicate filtering now applies the control sample's own duplicate threshold to control reads.
• Corrected lower-tail Poisson probabilities for very large expected counts.
• FRAG files may end with a record that has no trailing newline.
• Truncated BAM alignment records are rejected as invalid input.
• Improved reliability and cleanup when peak-calling workers fail or multiple operations run concurrently.
• Documented the Poisson tail convention used for reported p-values.

Compatibility

MACS3 3.1.0 requires Python 3.12 or later, as did MACS3 3.0.5.

Acknowledgements

Thanks to Jacob Schreiber for the callpeak auto-optimize work (PR #755, #756); Cindy Krafft for the duplicate-threshold and Poisson fixes and p-value documentation (PR #753, #777, #789); Michael R. Crusoe for spelling fixes (PR #754); and Apoorva Garg for the native aarch64 CI contribution (PR #793).