AlphaJudge 1.2.0

AlphaJudge 1.2.0

Added

• iLIS interface score (AFM-LIS; Kim et al., github.com/flyark/AFM-LIS): iLIS = sqrt(LIS * cLIS), where cLIS is the LIS PAE transform averaged only over residue pairs in direct contact (CB-else-CA within contact_thresh, default 8 Å). Adds Interface.clis()/ilis() and emits interface_cLIS / interface_iLIS next to interface_LIS in the per-run CSV. Validated against the official AFM-LIS lis.py to 4 decimal places.

Changed

• Percentile sliders and meta-score are now calibrated on positive (interacting) benchmark pairs only, so predictions rank against the distribution of real interfaces. Per-feature AUROC unchanged (monotonic); production interface_meta_score AUROC = 0.878. Reproduce deciles with scripts/freeze_metascore_quantiles.py --label-filter positive|negative|all.
• Removed the report slider poly-line; the Meta marker is recomputed from the current calibration for consistency.

Tests: 34 passed, 6 skipped; CI green on Python 3.10–3.13.

1.1.0

First release of the validation-report feature set and the interface meta-score.

Added

• RCSB / wwPDB-style PDF validation report (--report / --no-report). Writes a per-run report.pdf next to interfaces.csv: a percentile slider panel for every detected interface, a per-interface table for multimers, an AFDB-style PAE panel, and a complex-level confidence + PAE evidence page. On by default for single-run scoring; off when --summary is used so benchmark aggregations stay fast.
• --aggregate_report PDF: a multi-page cohort validation PDF built from the --summary CSV — meta-score histogram, top-N interfaces table, one slider page per interface ranked by meta-score, then a "Per-complex evidence" section (top-N capped by distinct complex). Requires --summary.
• Interface meta-score averaged percentiles score used to rank and summarise interfaces in the reports.

Changed

• The interface meta-score (and report) now use independent interface_hb (interface hydrogen bonds) instead of interface_area.
• Report visuals follow wwPDB/RCSB conventions
• Report generation is backend-agnostic: AF2, AF3, and Boltz-2 all flow through the same scoring path, so --report / --aggregate_report work for any mix.

Fixed

• --aggregate_report guards against a stale summary CSV instead of emitting an inconsistent report.

Full changelog: 1.0.2...1.1.0

1.0.2

What's Changed

• [codex] Add official AF3 layout and Boltz-2 parser support by @DimaMolod in #12

Full Changelog: 1.0.1...1.0.2

1.0.1

What's Changed

• Feature/performance improvements by @HenriettaHolze in #10

• Optimize memory by keeping PAE matrices as numpy arrays, credits to @HenriettaHolze
• A new flag to skip expensive biophysical scores computation. Thanks to @HenriettaHolze
• New flag to force recompute summary csv files
• New flag to skip saving PAE png plots
• New flag to change filenames of the output interfaces.csv

New Contributors

• @HenriettaHolze made their first contribution in #10

Full Changelog: 1.0.0...1.0.1

1.0.0

new code structure, more accurate biophysical scores.

What's Changed

• PR #8

Full Changelog: 0.3.0...1.0.0

AlphaJudge v0.3.0

The first release.