Skip to content

Share ligand preparation and add portable benchmark workflows - #11

Merged
th2ch-g merged 11 commits into
mainfrom
refactor/shared-ligand-preparation
Oct 13, 2026
Merged

th2ch-g merged 11 commits into
mainfrom
refactor/shared-ligand-preparation

Conversation

@th2ch-g

@th2ch-g th2ch-g commented Oct 12, 2026

Copy link
Copy Markdown
Contributor

Protenix and Chai now preserve native ligand reference coordinates and source chemistry for shared RGI force-field relaxation and stereo validation. This removes unconditional adapter-side conformer regeneration and keeps stereo repair in the shared preparation path.

  • Add portable prepare/run/evaluate workflows for conformer, distance, angle, RMSD, and custom restraints across Boltz-2, Protenix, AF3, Chai, OpenFold3, and ESMFold2. Include public dataset fetching, matched OFF/RGI jobs, runtime configuration, resume checks, software provenance, and CSV evaluation.
  • Validate Chai's converted MSA query before replacing prepared inputs, preserve valid inputs when conversion fails, and use the prepared MSA directory for both OFF and RGI arms.
  • Add adapter and benchmark regressions, usage documentation, and ignore rules for generated artifacts. Update the toolkit to 0.4.3 and the benchmark package to 0.1.1.

Validation: task lint passed (168 files already formatted), and git diff --check origin/main...HEAD passed. Full CPU/compiler suites, native predictor tests, and GPU inference were not rerun while opening this PR.

th2ch-g and others added 11 commits October 12, 2026 02:46
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Add matched MSA subsampling and native guidance arms, USalign and
MolProbity evaluation, confidence and ligand clash metrics, comparison
figures, and manuscript source-data exports. Use full RGI checkout names
and document external tool installation without changing package versions.

Co-authored-by: Codex <noreply@openai.com>
Prepare portable PLINDER receptor frames, report ligand pose RMSD after pocket alignment, and score E/Z with all heavy substituents. Cover incomplete pose inputs and atom-order invariance in the benchmark tests.

Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Co-authored-by: Codex <noreply@openai.com>
Preserve the reorganized README and restore Boltz-1 and Protenix v1 documentation.

Co-authored-by: Codex <noreply@openai.com>
@th2ch-g
th2ch-g merged commit 6a8bb6d into main Oct 13, 2026
3 checks passed
@th2ch-g
th2ch-g deleted the refactor/shared-ligand-preparation branch October 13, 2026 07:15
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