STRC Pharmacochaperone Phase 4 Plan — seven computational gates before any wet dollar

The Phase 0–3C screen on STRC Pharmacochaperone Virtual Screen E1659A ended with a 29-compound 2D+shape-fit shortlist and a pocket at centroid (9.52, −6.14, −39.83) Å with druggability 0.856. That is not yet enough to say “the hypothesis is validated”. This note defines the seven computational gates that must clear — with kill criteria — before any Enamine synthesis or explant assay is justified. All steps run on local compute against existing AF3 CIFs; total cost $0 (or \leq$ 500 if we rent an A100 for MD).

Docking target (reused across all phases)

Target CIF: job-ultramini-x-tmem145-full.cif (ARM repeat zone 1603-1749 intact, clinical construct)
WT reference CIF: job4-wildtype.cif (Phase 2b pocket derivation)
E1659A mutant CIF: job3-mutant.cif (Phase 1B loop displacement origin)
Docking box: 18 × 18 × 18 Å centred at (7.7, −5.4, −41.5) Å (merged subpocket centroid from Phase 2b, covers both top hits at (9.52, −6.14, −39.83) and (5.89, −4.68, −43.13))
Pharmacophore anchors: K1141-NZ (salt-bridge), D1140/D1173 (LRR face), F1646 (π-stack), loop 1642-1651 (cap target)
Positive control ligand: diflunisal (TTR pharmacochaperone; must rank in top-20)
Negative control ligands: glucose, urea, acetamide (polar, no aromatic, no COOH — must rank bottom decile)

The seven gates

#	Phase	Wall clock	Gate (advance if)	Kill criterion
1	4a — Pocket reproducibility	30 min	K1141 pocket present in ≥4/5 existing AF3 STRC CIFs with fpocket druggability ≥0.7	<4/5 → pocket is a one-job artifact; hypothesis dies
2	4b — Vina + GNINA real docking	2–3 d	≥3 of consensus top-20 better than diflunisal on Vina score ∧ GNINA CNNscore ≥0.7	<3 → shortlist is noise; pivot to ZINC22 15M tier
3	4c — WT control dock	1 d	Top-5 leads rank ≥2 kcal/mol worse on WT than on E1659A pocket	Leads bind WT ≥ E1659A → pocket is generic, not rescue-specific
4	4d — K1141A in-silico decoy	1 d	Carboxylate leads lose ≥2 kcal/mol when K1141→A	No loss → salt-bridge pharmacophore is false
5	4e — Off-target selectivity	hours	ΔG(K1141 pocket) − ΔG(best off-target Ultra-Mini pocket) ≥ 2 kcal/mol for ≥3 of top-5	No selectivity → not a pharmacochaperone, just a COOH magnet
6	4f — Interface-rescue docking	1 d	MM-GBSA on Ultra-Mini × TMEM145 interface with top-3 pre-bound: recovers ≥30% of the 8.4 kcal/mol binding-energy gap	No recovery → pocket binding does not translate to rescue; kill
7	5 — GROMACS MD stability	days	50 ns × 3 replicates: ligand heavy-atom RMSD <3 Å; K1141-NZ contact >60% frames; loop 1642-1651 RMSD to WT <2 Å; MM-PBSA ≤ −6 kcal/mol	All drift / MM-PBSA > −5 → no stable rescue pose; kill

Phase 6 (OpenFE FEP+) explicitly deferred — requires rented H100 (~$2-5k). The seven gates above are sufficient to decide “wet-lab green-light” without FEP+.

Libraries (all free)

DrugBank FDA subset (~2,500, academic-free)
DSi-Poised Library (~2,000, Diamond, free)
ZINC22 in-stock carboxylate tranche (~40,000, free)
Top-5 shortlist + diflunisal positive + 3 negative decoys from STRC Pharmacochaperone Virtual Screen Ranked Leads — mandatory fixed rows in every docking run

No Enamine on-demand until gates 1–7 pass.

Ligand preparation protocol

3D conformer generation: RDKit ETKDGv3 + MMFF94s minimisation
Protonation at pH 7.4: Open Babel obabel -p 7.4
Ionisation tautomers enumerated; up to 3 per molecule retained
Output: PDBQT via Meeko or AutoDockTools

Decision tree after all seven gates

7-pass → [[STRC Pharmacochaperone Virtual Screen E1659A]] S-tier confirmed; wet-lab authorised
6-pass + 7-fail → pose unstable; redesign scaffold, re-enter Phase 4b
any-of-4c/4d/4e fails → pocket/pharmacophore wrong; hypothesis D-tier (killed)
any-of-4a/4b/4f fails → hypothesis D-tier (killed); compute redirects to Mini-STRC Phase 4 or Strategy B LNP tropism scan

Files / Models

Scripts live in ~/STRC/models/:

pharmacochaperone_phase4a_pocket_reproducibility.py / .json — fpocket across 5 CIFs
pharmacochaperone_phase4b_vina_gnina_screen.py / .json — ~45k-compound real dock
pharmacochaperone_phase4c_wt_decoy.py / .json — WT control redock
pharmacochaperone_phase4d_k1141a_decoy.py / .json — in-silico mutant decoy
pharmacochaperone_phase4e_offtarget_selectivity.py / .json — Ultra-Mini surface scan
pharmacochaperone_phase4f_interface_rescue.py / .json — MM-GBSA on TMEM145 complex
pharmacochaperone_phase5_md.py / .json — GROMACS 50 ns × 3 replicates

Ranking delta

STRC Pharmacochaperone Virtual Screen E1659A: no tier change. Stays S-tier. This note is the execution plan, not a proof. Evidence depth unchanged until Phase 4a runs. Next step column in STRC Hypothesis Ranking updated from “wet-lab bind assay on salicylic acid / indole-3-acetic lead pair” → “run Phase 4a pocket reproducibility gate” (ordering blocked until computational validation closes).
STRC Mini-STRC Single-Vector Hypothesis: no change (Ultra-Mini × TMEM145 CIF is reused as docking target; this note consumes but does not alter Mini-STRC evidence).
All other S/A/B/C tier hypotheses: no change.

Connections

[part-of] STRC Pharmacochaperone Virtual Screen E1659A
[see-also] STRC Pharmacochaperone K1141 Fragment Pocket — pocket geometry this plan docks into
[see-also] STRC Pharmacophore Model K1141 Pocket — anchor triangle this plan validates
[see-also] STRC Pharmacochaperone Virtual Screen Ranked Leads — top-5 shortlist under gate test
[see-also] STRC Pharmacochaperone Loop 1642-1651 Target — loop-rescue geometry this plan measures
[see-also] STRC Ultra-Mini Full-Length TMEM145 AF3 — docking target CIF for gate 4f
[see-also] STRC Electrostatic Analysis E1659A — 8.4 kcal/mol gap this plan attempts to recover
[see-also] STRC Hypothesis Ranking — ranking register that advances/kills on gate outcome
[applies] Misha — maternal E1659A is the specific mutation this plan validates rescue for

STRC Research

Explorer

STRC Pharmacochaperone Phase 4 Plan