STRC h01 Phase 7H AF3 Protein-Ligand Validation 2026-04-24

STRC h01 Phase 7H AF3 Protein-Ligand Validation

12 AlphaFold Server protein-ligand jobs queued to independently validate the K1141 pocket + WT-vs-E1659A binding-mode difference. Mini-STRC (residues 1075-1775, 701 aa) × 6 NSAID-class reference ligands (all PDB CCD codes verified via RCSB text_chem search 2026-04-24) × 2 backgrounds (WT + E1659A). Jobs are model-independent from Vina — success criterion is AF3 placing the acidic head of each ligand within 4 Å of K1141 Cα, without pose hint. Verdict pending 6-12 h AFS queue.

Problem

After Phase 4c-v3b + 4d falsified the static-Vina form of the K1141-salt-bridge pharmacochaperone claim (STRC h01 Phase 4c-v3b WT Decoy on v3b YELLOW 2026-04-24, STRC h01 Phase 4d K1141A Double-Mutant Decoy 2026-04-24), a Nobel-tier reviewer will ask: “is the K1141 pocket itself even real, or is it a Vina-pipeline artefact of the AF3 static structure?”

The question is non-trivial because Phase 2 / 2b / 3a / 3b pipelines all treat the pocket as a given. AF3 placed the Mini-STRC + TMEM145 complex, SiteMap-style druggability scoring picked the K1141 cluster as highest, and every subsequent docking phase ran in that box. If AF3 independently places an NSAID ligand ANYWHERE ELSE on the 701-aa Mini-STRC surface when given only (sequence, ligand CCD code) as input, the pocket hypothesis has a problem.

Conversely: if AF3 (a model trained on crystal-bound protein-ligand complexes, with no Vina data in its training set) places NSAIDs at or adjacent to K1141 — that is independent evidence that the pocket exists in real biomolecular space.

Method

• Tool: AlphaFold 3 Server v1 alphafoldserver dialect (AlphaFold 3 Server tool card).
• Script: [[af3_jobs_2026-04-24_phase7h_builder.py]] (new 2026-04-24).
• Construct: Mini-STRC (residues 1075-1775 of UniProt Q7RTU9, 701 aa) — matches STRC AF3 Truncation Boundary Sweep optimum (pTM 0.86-0.87), contains all K1141 pocket residues (K1141, D1140, D1173, F1646, E1659).
  • Mini-STRC numbering: K1141 → Mini-pos 67; E1659 → Mini-pos 585.
• Mutant: E→A at Mini-pos 585.
• Ligands (PDB CCD codes verified via RCSB text_chem contains_phrase query 2026-04-24):

  CCD	name	context
  1FL	diflunisal	Phase 4b baseline carboxylate
  3MI	tafamidis	Phase 4h Tier-1 positive (FDA-approved TTR stabiliser)
  NFL	niflumic acid	Phase 3c v2 top hit (best-Kd fenamic)
  FLF	flufenamic acid	Phase 4c legacy lead + Phase 6c COX-1 precedent
  FHI	iododiflunisal	Phase 4h Tier-1 iodinated diflunisal
  ID8	mefenamic acid	Phase 6c COX-2 5IKR reference

• 6 ligands × 2 backgrounds (WT, E1659A) = 12 jobs. Each ≈ 730 tokens (701 aa protein + ~30 heavy-atom ligand) — well under 5,000-token AF3 limit.
• Model seed: 42 (consistent with prior STRC AF3 runs for seed-reproducibility).

Limitation — novel Phase 4h combinatorial candidates not submittable

AF3 Server only accepts PDB Chemical Component Dictionary (CCD) codes for ligands — SMILES is not currently supported via the web UI. The 29 v3b YELLOW + 30 v4 YELLOW combinatorial compounds and the novel Phase 4h Tier-1 candidates (benzox-3OH-4F-COOH #15, benzimid-3OMe-COOH #27, indole-3OMe #17) are not in the CCD and therefore cannot be submitted.

This is a real scope limitation: Phase 7H tests the pocket and the reference-ligand binding mode, not the novel combinatorial winners. To AF3-validate the novel candidates, a local AF3 installation (H100 GPU + MSA data) or parallel Boltz-2 run would be needed — queued for a separate phase.

Jobs queued

af3_jobs_2026-04-24_phase7h/
  MANIFEST.json
  af3_jobs_sequences.fasta
  h01_phase7h__mini_wt__{diflunisal,tafamidis,niflumic-acid,flufenamic-acid,iododiflunisal,mefenamic-acid}__{1FL,3MI,NFL,FLF,FHI,ID8}.json
  h01_phase7h__mini_e1659a__{...}__{...}.json

Submission step (requires Google auth in browser; per AlphaFold 3 Server skill Method 1): navigate to alphafoldserver.com, log in, click “Upload JSON” for each file, click “Submit 1 job”. 12 jobs, well under the 20/day limit. Expected delivery 6-12 h wall.

Success criteria (verdict, pending delivery)

Per-job metrics from AF3 *_summary_confidences_*.json:

1. Pocket placement. Extract ligand heavy-atom centroid and compute distance to K1141 Cα (Mini-pos 67). Success: ≤ 6 Å centroid-to-Cα for ≥ 4 of 6 WT-background ligands.
2. Ligand confidence. AF3 per-ligand pLDDT > 50 (low = model uncertain about placement). Success: median pLDDT ≥ 60.
3. WT-vs-mutant difference. ΔG proxy — ipTM on the ligand interface — must differ by ≥ 0.1 between WT and E1659A for ≥ 2 of 6 ligands. A ZERO difference across all 6 would confirm Phase 4c-v3b finding that AF3 doesn’t resolve mutant pocket. A clear difference would RESCUE the static claim.
4. Acidic head orientation. For carboxylate-class ligands (1FL, NFL, FLF, FHI, ID8), the COO⁻ group should point toward K1141 ε-NH3+ in the AF3 pose. Measured as COO-C / K1141-NZ distance ≤ 5 Å.

Ranking delta

• Hypothesis h01: tier A held, mech 2 held until AF3 delivery
• Provisional next-step update pending outcomes:
  • If AF3 places NSAIDs at K1141 pocket AND resolves WT-vs-mut difference: mech 2 → 3 (pocket validated independently; dynamic claim still open)
  • If AF3 places NSAIDs at K1141 but no WT-vs-mut difference: mech 2 held; dynamic claim remains only route
  • If AF3 places NSAIDs elsewhere on Mini-STRC surface: mech 2 → 1 (pocket hypothesis challenged); pivot to cryptic-pocket search on Phase 5d trajectory + h03 alternative
• This note will be updated with final Ranking delta after AF3 results land.

Connections

• [part-of] h01 hub
• [see-also] STRC h01 Phase 4c-v3b WT Decoy on v3b YELLOW 2026-04-24
• [see-also] STRC h01 Phase 4d K1141A Double-Mutant Decoy 2026-04-24
• [see-also] STRC AF3 Static Pocket Blindness to Loop Dynamics
• [see-also] STRC Pharmacochaperone K1141 Fragment Pocket
• [source] AlphaFold 3 Server