This runbook documents the current local-only protein design loop. It stays explicit about what is deterministic, what is mock/demo, and what files to open when you want to continue the next run.
- Start from
protein_loop.pyin the repo root. - The runner loads the current demo backbone, sequence, MSA, and paired-MSA artifacts.
- The weighting pipeline scores the homolog family and updates heuristic state.
- The runner writes the summary, comparison, next command, rationale, review, and session index.
- The root status page and runbook point back to the live handoff files.
- mode:
mixed - status:
MOCK / DEMO - backbone:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\mixed_fold_backbone.pdb - folded:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\mixed_fold_folded.pdb - sequence file:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\sequences.fa - msa:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\openfold3_msa_mock.a3m - paired msa:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\openfold3_paired_msa_mock.csv
- summary:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\protein_loop_run_summary.json - comparison:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\openfold3_run_comparison.json - next command:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\openfold3_next_run_command.txt - rationale:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\openfold3_next_run_rationale.txt - review:
C:\Users\Zhane\Documents\New project\zrt-bionemo\outputs\openfold3_run_review.md - session index:
outputs/openfold3_session_index.md - root status:
protein_loop_status.md
- next rule:
exploit-neff - next temperature:
0.1951 - next diversity strength:
0.633 - next neighbor identity:
0.9
- latest vs previous msa neff delta:
0.106226 - latest vs previous paired neff delta:
0.120243 - latest vs best msa neff delta:
-0.84538 - latest vs best paired neff delta:
-1.127252
MOCK/DEMOonly.- No hidden autonomy.
- Explicit overrides always win over seeded defaults.
- The feedback loop is deterministic heuristic update, not learned self-training.
- External BioNeMo calls remain separate from the local core.