STRC h01 Phase 9 — Orthogonal Cross-Checks Plan

Plan note. Five tracks of independent in-silico validation prepared as runnable skeletons. No production runs yet — this document defines the menu, the verifiable pass criteria, and the smoke-test rules so the production runs can fire tomorrow/day-after.

Why this phase exists

The h01 differential (mutant E1659A vs WT for chaperone-pocket binding) currently rests on three pillars:

  1. Vina / Boltz-2 docking consensus (Phase 5e, 5q) — known artifacts: pose-search bias, force-field charge approximations, scoring-function calibration error.
  2. tauRAMD residence times (Phase 5m, 5q) — known artifacts: forced-pull non-equilibrium, force constant choice, mass scaling.
  3. APBS continuum electrostatics (Phase 5j, 5k, 8g-v2) — known artifacts: PB box geometry, dielectric assumption, no explicit waters/ions in pocket.

Each pillar can be wrong in a way that conspires with the other two (shared force-field family, shared docking-derived starting poses). Reviewers — and an honest internal audit — need cross-checks that use different physics (FEP, PMF, QM) or different model families (independent AI co-folders) to confirm the differential survives.

Tracks (skeletons exist, awaiting production runs)

#MethodTestsSkeleton
9aFEP/TI alchemical ΔΔGabsolute/relative binding free energy via λ-mutation E→ASTRC h01 Phase 9a FEP-TI E1659A Alchemical 2026-04-27 SKELETON
9bHolo MD + MM-PBSAlong-trajectory complex stability + isolated electrostatic ΔGSTRC h01 Phase 9b Holo MD MM-PBSA 2026-04-27 SKELETON
9cUmbrella sampling PMFequilibrium dissociation barrier (replaces forced-pull tauRAMD)STRC h01 Phase 9c Umbrella Sampling PMF 2026-04-27 SKELETON
9dQM/MM single-point + optquantum treatment of charged-residue pocket; polarization/charge-transferSTRC h01 Phase 9d QM-MM Pocket 2026-04-27 SKELETON
9eBlind neural dockingindependent AI co-fold (AF3 / DiffDock / Boltz-2 / Uni-Mol)STRC h01 Phase 9e Blind Neural Docking AF3 DiffDock 2026-04-27 SKELETON

Pass criteria for the phase as a whole

  • PRIMARY: ≥3 of 5 tracks reproduce the sign of the existing differential (mutant binds better than WT).
  • SECONDARY: at least one of {9a, 9c} produces a quantitative ΔΔG estimate within 2 kcal/mol of the MM-PBSA value (9b).
  • TERTIARY: 9e blind co-fold places lead ligand in the K1141 pocket for ≥2 of 4 AI models without seeded coordinates.
  • FAIL state: ≥3 tracks show no differential or opposite sign → freeze h01 wet-lab planning, return to mechanism re-derivation.

Smoke-test rule (applies to every track)

Each skeleton specifies a 1-day smoke variant that runs on existing AF3 / Phase 5d MD / Phase 5q docked assets — no re-equilibration. Smoke output is a binary go/no-go on the production protocol, not a scientific result. Production fires only after smoke passes.

Candidate extras — menu of further cross-checks (not skeletoned yet)

If the 5 main tracks come back ambiguous or a reviewer asks for more, these are the next-tier options ordered by orthogonality / ease:

ExtraWhat it addsCost
Metadynamics / well-tempered MetaDfaster PMF than umbrella; orthogonal CV biasingmedium
GaMD (Gaussian-Accelerated MD)enhanced sampling without CV definitionlow-medium
WaterMap / GIST / 3D-RISMwater-displacement thermodynamics in pocketlow
Markov State Models from existing MDkinetic decomposition orthogonal to single-trajectory tauRAMDlow (post-process)
LIE (Linear Interaction Energy)cheap end-state alternative to FEPvery low
MM-GBSA per-residue decompositionresidue-specific energy contribution to differentialvery low (post-process)
FTMap / SILCS / MixMD solvent probeorthogonal pocket druggability maplow
DCCM / network analysis on MDallosteric signal from E1659 to K1141 pocketlow (post-process)
Force-field cross (AMBER ↔ CHARMM ↔ OpenFF)robustness to FF familymedium
Water-model cross (TIP3P ↔ OPC ↔ TIP4P-Ew)robustness to water parameterizationmedium
Charge-isosteric mutational panel (E1659D / E1659K / E1659G)does mechanism follow charge?medium
ESMFold / OmegaFold structural prior comparisonorthogonal structure to AF3 starting modelslow

These are documented here for future scoping, not yet skeletoned as phase notes. Promote to a Phase 9f-9q skeleton only when the corresponding decision becomes load-bearing.

Tooling readiness — pre-flight matrix (to fill in before launching)

TrackSoftwareLicense?Hardware?Existing assets reusable?Smoke wall-time
9aGROMACS-FEP / AMBER pmemd / pmxTBDGPU (M5 Max MPS or remote A100)Phase 5d holo trajectoryTBD — fill before run
9bGROMACS + gmx_MMPBSA / AmberToolsinstalled locally; input-blockedGPU for production, local CPU OK for setupPhase 5q E1659A OpenMM assets only; matched WT holo missingsmoke blocked until matched WT/MUT topologies exist
9cGROMACS pull + gmx whamTBDGPUPhase 5d starting frameTBD
9dORCA + AMBER / CP2K / ChemShellTBDCPU-heavy (cluster?)Phase 5e shortlisted posesTBD
9eAF3 server + DiffDock + Boltz-2 + Uni-Molalready integrated for AF3/Boltz-2local + AF3 serverPhase 5q shortlist + ligand SMILESTBD

Each row gets filled inside its respective skeleton before that track fires.

Status

  • 2026-04-27 00:30 GMT+8 — Phase 9 menu opened; 5 skeleton phase notes created; tooling matrix unfilled. No production runs yet.
  • 2026-04-27 11:29 GMT+8 — Phase 9b-lite residue-field smoke delivered from existing matched WT/MUT MD snapshots: broad shells support E1659A>WT sign (18 A delta +0.181 q/A, p=1.54e-10), with 9 A inner-shell caveat. This is supportive smoke, not one of the 5 production track verdicts. → STRC h01 Phase 9b-lite Residue Field Decomposition 2026-04-27
  • 2026-04-27 11:45 GMT+8 — Phase 9x v5.3 alpha sensitivity delivered: adamantyl_acylsulfonamide_SO2Me_-Cl stays rank-1 for alpha 0.5-2.0 across Boltz-2, Vina, tauRAMD, and ADMET rank axes. Supportive lead-committee robustness; not a production track verdict. → STRC h01 Phase 9x v5.3 Lead Committee Alpha Sensitivity 2026-04-27
  • 2026-04-27 11:54 GMT+8 — Phase 9b MM-PBSA tooling preflight delivered: GROMACS / gmx_MMPBSA / AmberTools callable, but WT-vs-E1659A MM-PBSA is input-blocked. No native GROMACS/Amber h01 topologies exist; Phase 5q has only E1659A OpenMM holo assets; post-hoc OpenMM Amber conversion is invalid due to missing bond types. → STRC h01 Phase 9b MM-PBSA Tooling Preflight 2026-04-27

Ranking delta

  • h01: no change. This is a planning + cross-check infrastructure phase. Tier/mech/deliv/misha_fit/next_step do not move until at least one of 9a-9e returns a verdict. Recording this so the no-change is auditable.

Connections