Skip to content

Release 1.2.0: iLIS interface score + positives-only slider calibration#18

Merged
DimaMolod merged 1 commit into
mainfrom
release-1.2.0
Jun 15, 2026
Merged

Release 1.2.0: iLIS interface score + positives-only slider calibration#18
DimaMolod merged 1 commit into
mainfrom
release-1.2.0

Conversation

@DimaMolod

Copy link
Copy Markdown
Collaborator

Bumps alphajudge to 1.2.0 and adds a CHANGELOG entry for the two functional commits already on main (3f5ec10, 45dc059).

What's in this release

  • iLIS interface score (AFM-LIS; Kim et al.): iLIS = sqrt(LIS * cLIS), where cLIS is the LIS PAE transform averaged only over residue pairs in direct contact. Emits interface_cLIS / interface_iLIS in the per-run CSV; validated against the official AFM-LIS lis.py to 4 decimals.
  • Positives-only calibration: percentile sliders and meta-score are now calibrated on interacting (positive) benchmark pairs only, so predictions rank against the distribution of real interfaces. Per-feature AUROC unchanged (monotonic); production interface_meta_score AUROC = 0.878. The report slider poly-line is dropped.

Validation

  • Full test suite on this branch: 34 passed, 6 skipped (pytest -n auto --dist loadfile, BLAS threads capped).

🤖 Generated with Claude Code

- Add iLIS (AFM-LIS) interface score: sqrt(LIS * cLIS), emitting
  interface_cLIS / interface_iLIS in the per-run CSV.
- Calibrate percentile sliders and meta-score on positive (interacting)
  benchmark pairs only; drop the report slider poly-line.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@DimaMolod DimaMolod merged commit a1b022e into main Jun 15, 2026
12 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant