STRC h01 Phase 5c Cryptic Pocket Analysis 2026-04-23
TL;DR. K1141 pocket is structurally stable across 2 ns MD (Cα RMSF 0.62 Å, half the global mean 1.23 Å; local void volume 719-850 ų across 20 frames, ~15% variance). No alternative cavity exceeds 152 ų (vs K1141’s ~750 ų). Phase 5b RED-LIGHT is chemistry-limited, not site-limited. Phase 3c v2 expanded virtual screen is correctly targeted at K1141 — with ensemble-receptor targeting to exploit the 20 MD conformers. This rules out the “maybe K1141 is the wrong pocket” hypothesis as the explanation for the RED-LIGHT shortlist.
Context
Phase 5b RED-LIGHT (all leads f_PC < 0.10, see STRC h01 Phase 5 MD Ensemble Rescoring 2026-04-23) could reflect either:
- (a) chemistry-limited: current 5-lead shortlist + diflunisal are inadequate; K1141 is druggable but we haven’t found the right scaffold yet → Phase 3c v2 expanded screen is the answer
- (b) site-limited: K1141 pocket as modelled in the static Ultra-Mini × TMEM145 CIF is non-physiological (opens differently in solution, collapses, or is adjacent to a better cryptic site) → Phase 3c v2 needs retargeting
Phase 5c decides between (a) and (b).
Method
fpocket 4.0 from Homebrew has a qhull/Voronoi incompatibility that raises QH6047 upper-Delaunay on every input (tested on stripped heavy-atom PDBs with jitter; reproducible). Rather than yak-shave a source build, we wrote a custom grid-based cavity analysis + K1141 stability tracker (pharmacochaperone_phase5c_cryptic_pocket_detection.py):
- Snapshot loading: 20 production-MD snapshots from Phase 5a, H-stripped to 5411 heavy atoms / 701 residues each. Kabsch-aligned on Cα to snap_000 as reference.
- K1141 pocket stability:
- Identify “pocket-lining residues” as those with any heavy atom within 6 Å of K1141 pocket centre
(-22.027, -18.547, 2.215)(Phase 4b box centre) in snap_000. - Per-residue Cα RMSF across aligned frames.
- Local void volume per frame: 8 Å sphere around K1141_CENTRE, 1.5 Å voxel grid, count voxels with no heavy atom within 2.0 Å (proxy for pocket openness).
- Identify “pocket-lining residues” as those with any heavy atom within 6 Å of K1141 pocket centre
- Global cavity scan (mid-trajectory snap_010):
- 1.5 Å voxel grid over protein bbox + 4 Å pad (226 k voxels).
- Candidate voxel: 2.6 Å < d(nearest heavy atom) < 4.5 Å.
- Burial test: cast 26 cube+face+edge-diagonal rays, step 1.0 Å × 6 times, hit threshold 1.8 Å. Voxel is “buried” iff ≥ 16/26 rays hit protein (≥ 60% encasement).
- Cluster buried voxels via
scipy.ndimage.label(3D 6-connected components). - Report clusters ≥ 20 voxels (~70 ų) with volume, centroid, distance to K1141.
Results
K1141 stability — STABLE
| Metric | Value | Global baseline | Interpretation |
|---|---|---|---|
| K1141 pocket Cα RMSF | 0.62 Å | 1.23 Å | Pocket residues are 2× more rigid than average |
| Pocket-lining residues (static) | 9 residues: 66, 67, 68, 538, 541, 542, 570, 571, 572 | — | 3 loop segments converge on K1141 |
| Local void volume (20 frames) | 719.0 – 850.5 ų (mean 789.8 ų) | — | 15% CV, pocket does not close |
The K1141 pocket residues have RMSF 0.62 Å — half the global mean of 1.23 Å. Combined with void volume holding steady at 720-850 ų across the trajectory, the pocket neither collapses nor opens into a different geometry during MD.
Alt cavity scan — no competing druggable site
Top 5 alt cavities on snap_010 (distance > 6 Å from K1141, buried ≥ 60% of directions):
| Rank | Volume (ų) | Centre (x, y, z) | Distance to K1141 (Å) |
|---|---|---|---|
| 1 | 151.9 | (−2.1, −25.3, +24.7) | 30.8 |
| 2 | 128.2 | (−1.9, −13.4, +22.5) | 29.0 |
| 3 | 101.2 | (−12.2, −24.6, +20.0) | 21.2 |
| 4 | 101.2 | (+20.2, −28.4, +14.9) | 45.1 |
| 5 | 91.1 | (−15.0, −13.8, +18.8) | 18.7 |
None approach K1141’s local void volume (~750 ų within an 8 Å sphere — not directly comparable, but the 8 Å sphere void is ≫ alt cavity voxel count even after correction). Top alt cavity is 152 ų at 31 Å from K1141 — spatially unrelated, small, and likely a small surface concavity rather than a drug-grade buried pocket.
Verdict: Phase 5b red-light is chemistry-limited
The structural premise of h01 — that K1141 is a druggable E1659A-proximal pharmacochaperone pocket — survives MD scrutiny. The Phase 4b/5b shortlist (naphthalene-2-COOH, cyclopropane-phenyl-COOH, indole-3-acetic-acid, salicylic, nicotinic + diflunisal positive) delivers f_PC < 0.10 against this pocket because the chemistry is wrong, not because the pocket is wrong.
Consequences
- Phase 3c v2 plan confirmed: expanded virtual screen against K1141 with ensemble receptor docking (use all 20 Phase 5a snapshots, not a single static structure) is the correct next compute move.
- Ensemble docking implementation: ~5 representative conformers selected by k-means over pocket backbone coords; screen library against each; rank by mean ensemble ΔG.
- No retargeting: no large cryptic cavity justified parallel screening at a different site. Future PCs should bind K1141 as designed.
- Bonus hedge: keep a 5-10% compute budget for the three nearest alt cavities (rank 3, 5 within 20 Å of K1141) in case Phase 3c v2 comes up empty — they’re too small to be primary targets but could host allosteric fragment hits.
What this doesn’t prove
- 2 ns is short. A 20-100 ns trajectory could still find rare cryptic pocket openings that we missed. Cheap extension: Phase 5a v2 runs 20 ns on the Mac (~8 hours wall) and re-runs 5c; defer until Phase 3c v2 delivers/fails.
- Our custom cavity detector is cruder than fpocket/P2Rank. It captures pocket volume and centroid correctly but lacks fpocket’s druggability score. If Phase 3c v2 returns a weak green, worth a P2Rank build to double-check nothing druggable was missed.
- Pocket stability in apo (unliganded) MD is a lower bound; ligand-induced fit could reveal additional binding modes. Already implicitly tested in Phase 5b ensemble re-docking, which showed no induced-fit rescue — consistent with stable-pocket red-light.
Ranking delta
- #1 (h01) Pharmacochaperone E1659A: A held. Mech 3 held, Deliv 4 held, Misha-fit 4 held. Phase 5c confirms K1141 as the correct site for Phase 3c v2; removes the “site-limited” hypothesis from the RED-LIGHT interpretation. Phase 3c v2 priority re-confirmed — expanded chem-space screen (DrugBank FDA + ZINC22 carboxylate tranche + fragment-based) with ensemble receptor targeting.
Connections
- STRC h01 Phase 5 MD Ensemble Rescoring 2026-04-23 — RED-LIGHT explained as chemistry-limited not site-limited
- STRC Pharmacochaperone K1141 Fragment Pocket — pocket characterisation confirmed in MD
[applies]STRC Pharmacochaperone Virtual Screen E1659A — Phase 3c v2 is the next step[see-also]Misha Compound-Het Therapy Stack Model — h01 path to NORMAL remains open but chem-space-gated[part-of]STRC Hypothesis Ranking Log[about]Misha