Multi-agent doer↔verifier audit history. Every iteration here was driven by a critical issue surfaced by 4 specialised review agents (validator, code reviewer, scientific officer, structural bioinformatician) running in parallel against the previous version. New agents reading this file should consult the round-N report files in reviews_vN/ for the verbatim verdicts.
Status: under audit. 4 reviewers running.
Built: 10 homology models of TYMS using Modeller 10.8 against 3 templates with 30–95 % sequence identity (educational, not 100 %), validated locally with Ramachandran φ/ψ + DOPE per-residue + RMSD vs crystal, plus a manual SAVES-upload procedure.
10_modeller/06_validation/SAVES_MANUAL.md because SAVES has no programmatic API.Verdict (4 reviewers): Validator 10/10 PASS · Code review v4 punch list closed · Scientific officer SHIP · Structural bioinformatician CONDITIONAL PASS (3 reporting-only items, all baked into report_FINAL.docx).
Hard fix in this round: in-place reprotonation of the original 1HVY crystal cofactor coordinates (0.000 Å heavy-atom drift, 0 protein clashes). v4 had Kabsch-aligned the CCD-ideal D16 onto the bound conformer, producing 2.71 Å heavy-atom drift plus a 1.95 Å clash to PHE 80 CD2 — that placement artefact drove the v4 “cofactor expels dUMP” interpretation, which v5 dissolved. WT holo recovered to top affinity −8.25 kcal/mol with named-RMSD 0.999 Å vs crystal dUMP.
Headline: Rigid-receptor AutoDock Vina with AD4 partial charges and the physically correct (net −2) raltitrexed cofactor cannot resolve TYMS active-site point mutants at the kcal/mol scale. Largest holo Δ Vina = +0.77 kcal/mol (R215A_N226A) — well below Vina’s ±0.85 noise floor. Null-result methodology paper.
Critical issue surfaced: v3’s cofactor “pH 7.4 reprotonation” was a no-op (output file byte-identical to v2). Holo electrostatics still wrong → C195A negative Δ was an artefact, not biology.
Fix in v4: RDKit reprotonation from the PDB Chemical Component Dictionary’s ideal SDF, bond-order assignment, deprotonation of both carboxylates, MD5-inequality assertion, plus atom-name preservation through the obabel↔Vina round-trip. v4 closed the v3 punch list — and exposed the deeper placement-artefact issue that round 5 fixed.
Critical issues surfaced:
get_strain() returned 1e30 in headless mode, so v2’s rotamer minimisation was a no-op (every mutant used rotamer 0).Fix in v3: charge waterfall (obabel Gasteiger → meeko → pdb2pqr) with max|q| > 0.05 gate, multi-seed WT holo with affinity-based selection, sculpt rotamer minimisation, positive=destabilising sign convention with mean_topk = mean(top min(3,n)).
Critical issues surfaced (the v1 audit):
P0CG53 is yeast polyubiquitin, P11849 is a T4-phage photosystem-II-family protein, P04996 is an L. casei membrane protein, etc.). Only human + mouse were real TYMS. JSD scores were noise.Fix in v2: real TYMS panel of 10 verified orthologs across 5+ kingdoms; A+B dimer kept; CME43 re-mutated to native CYS in place; G217W dropped on the heavy-atom clash check; both apo and holo dockings produced.
Initial pipeline. Stages 1–9 ran end-to-end against human TYMS (P04818) + dUMP from PDB 1HVY. The 4 reviewers immediately surfaced the critical MSA + dimer + G217W + CME43 issues that v2 had to fix.
| Round | Folder |
|---|---|
| 1 (v1) | reviews/ |
| 2 (v2) | reviews_v2/ |
| 3 (v3) | reviews_v3/ |
| 4 (v4) | reviews_v4/ |
| 5 (v5) | reviews_v5/ |
| 6 (Phase 6, in progress) | reviews_phase6/ (will be added when reviewers complete) |