STRC Research

STRC h01 Phase 8 v5 Library Coulomb-Aware Design 2026-04-24

8 min read

STRC h01 Phase 8 — v5 Library Design with Coulomb-Aware Ranking

Phase 5k Stage B showed that Vina’s v3b ranking is nearly orthogonal to ensemble Coulomb ranking (Vina top-1 is only #3 by Coulomb; top-1 Coulomb was Vina rank #8). This phase builds a 60-ligand v5 library around the top Coulomb scaffold (2-amino-quinoline-3) with 3 explicit design axes: (1) non-planar aliphatic/bicyclic tails for hERG mitigation, (2) acid-head bioisosteres for Coulomb optimization, (3) quinoline-C4 substituent tuning. After Meeko + AutoDock Vina 1.2.7 docking into static E1659A and pose-transplant ensemble APBS rescoring over 20 Phase 5d mutant snapshots, the combined objective (Vina_ΔG + APBS_formal_ensemble) produces a v5 top-20 with logP collapsed to 0.2-2.9 (vs v3b top-1 logP 5.27). Best combined lead v5__aq3__adamantyl__CONHOH__-Cl at −8.19 kcal/mol combined, 95% ensemble-positive. Phase 8 opens the medchem forward path for Misha with Coulomb-as-ranking-criterion.

Problem

Phase 5k Stage B (see STRC h01 Phase 5k Ensemble APBS on Phase 5d Mutant MD 2026-04-24) identified a critical design flaw in the v3b library: Vina ranking is nearly orthogonal to ensemble Coulomb ranking. The Vina v3b top-1 (nc__3-amino-benzofuran-2-COOH__1-naphthyl__CONHOH__-CF3) was only #3 by ensemble Coulomb; the top-1 Coulomb scaffold (2-amino-quinoline-3 × 1-naphthyl × CONHOH × CF3) was Vina rank #8. A medchem forward path on v3b’s Vina-only ranking would miss the molecule that best exploits the pharmacochaperone’s electrostatic asymmetry.

Phase 8 addresses this by (a) growing the v3b top-Coulomb scaffold with systematic chemistry variation, (b) introducing the Phase 5k ensemble-APBS score as a co-equal ranking criterion alongside Vina ΔG. Additional constraint: every non-planar scaffold replacement targets hERG mitigation per Phase 6c class liability (STRC h01 Phase 6c hERG Extension 2026-04-24, 21/28 v3b hit sub-10 µM hERG; planar bicyclic tails are the structural liability).

Method

v5 library generation (Phase 8a)

Base scaffold: 2-amino-quinoline-3 (SMILES template {R2}c1c(N{R1})nc2ccccc2c1{R3}, where R1 is the tail attached via the 2-amino nitrogen, R2 is the acid head at C3, R3 is the C4 quinoline substituent).

Design axes:

axisv3b variantsv5 variants introduced
R1 tailnaphthyl, biphenyl, 3-CF3-phenyl, 3,5-diMe-phenyl, 4-F-biphenyl, 4-Cl-phenyl (all planar)trans-decalinyl, adamantyl, bicyclo[2.2.2]oct-1-yl, norbornan-2-yl, spiro[4.4]non-1-yl, 4-tBu-cyclohexyl, 4-Me-4-OH-cyclohex, 2-tetrahydronaphth, 1-indanyl, 3,3-diMe-indanyl (all non-planar, sp3-rich)
R2 acid headCONHOH, CONHSO2Me, tetrazole, COOHCONHOH, tetrazole, COOH, CONHSO2Me + phosphonate, sulfamate, CONHOMe (Coulomb-expanded)
R3 quinoline C4CF3, F, Me, HCF3, F, Cl, OMe, CN, H (substituent-palette extended)

Enumeration + drug-likeness filter (MW < 500, logP < 5.5, HBA ≤ 10, HBD ≤ 5, rotB ≤ 10): 272 candidates → 60 trimmed by design-priority score (non-planar × CONHOH/tetrazole × CF3/F).

Library breakdown of the 60 finalists:

  • 48 planar-tail + 12 non-planar baseline; after trim prioritizing non-planar → heavy non-planar bias
  • Acid heads: 60 CONHOH (sensitivity sweep to isolate tail+R3 effects)
  • MW range: 310-460; logP range: −0.5 to +2.9
  • sp3 fraction: 0.30-0.52 (vs v3b top-1 sp3 ~0.10)

Script: pharmacochaperone_phase8_v5_library.py (new). Output: artifacts/phase8_v5_library/phase8_v5_library.{csv,sdf,json}.

v5 docking + ensemble-APBS rescoring (Phase 8b)

Pipeline per ligand (parallel-4, ~20 min wall for all 60):

  1. ETKDG 3D embed via RDKit + UFF optimization → single-molecule SDF
  2. Meeko mk_prepare_ligand.py → PDBQT with Gasteiger charges
  3. AutoDock Vina 1.2.7 dock into static E1659A: box centred (15.777, −44.660, 16.866), size 18³ (matched Phase 4c / 5j / 5k), exhaustiveness 16 (first-pass screen), num_modes 3, cpu 4 per worker
  4. Parse docked pose (MODEL 1): extract Vina ΔG + OA-atom centroid
  5. For each of 20 Phase 5k mutant APBS grids: transplant OA centroid via static-mutant→snap pocket-local Kabsch (residues 1126-1156), interpolate φ at transplanted position
  6. Compute ensemble median φ (with |φ|<20 spike filter), convert to formal-anion ΔG = −0.616 × φ
  7. Combined score = Vina_ΔG + α × ΔG_formal_ensemble_median (α = 1.0 unit-weighted)

Script: pharmacochaperone_phase8b_v5_dock_rescore.py (new). Output: artifacts/phase8b_v5_dock_rescore/phase8b_v5_dock_rescore.{csv,json}.

Results

60/60 ligands docked successfully; all had OA atoms for centroid extraction (CONHOH-only library).

v5 top-20 by combined score

rankligandVina (kcal)APBS ΔG_ens (kcal)combinedpos_fraclogP
1adamantyl__CONHOH__-Cl−6.66−1.53−8.1995%1.94
21-naphthyl__CONHOH__-F−6.17−1.86−8.02100%−0.06
3bicyclo[2.2.2]oct-1-yl__CONHOH__-Cl−5.89−1.58−7.4895%1.70
41-naphthyl__CONHOH__-Cl−7.04−0.41−7.4685%0.21
5adamantyl__CONHOH__-F−6.13−1.29−7.42100%1.67
6norbornan-2-yl__CONHOH__-CF3−5.74−1.56−7.30100%1.53
7bicyclo[2.2.2]oct-1-yl__CONHOH__-CN−6.04−1.24−7.2895%1.02
8adamantyl__CONHOH__-CN−6.20−1.02−7.22100%1.27
91-naphthyl__CONHOH__-CN−6.70−0.41−7.1185%−0.47
101-naphthyl__CONHOH__-H−6.73−0.32−7.0585%−0.36
111-indanyl__CONHOH__-CF3−5.98−1.06−7.04100%1.51
12biphenyl__CONHOH__-CF3−6.40−0.47−6.8795%2.88
13adamantyl__CONHOH__-CF3−6.50−0.37−6.8795%2.31
14spiro[4.4]non-1-yl__CONHOH__-CN−5.76−1.09−6.84100%1.41
153,5-diMe-phenyl__CONHOH__-CF3−6.00−0.79−6.79100%2.13
16adamantyl__CONHOH__-OMe−6.36−0.42−6.7895%1.35
17bicyclo[2.2.2]oct-1-yl__CONHOH__-CF3−6.25−0.49−6.74100%2.06
184-Me-4-OH-cyclohex__CONHOH__-CF3−5.22−1.46−6.68100%1.99
19bicyclo[2.2.2]oct-1-yl__CONHOH__-F−5.88−0.80−6.6790%1.43
203,3-diMe-indanyl__CONHOH__-Cl−5.71−0.96−6.66100%1.78

Observations

  1. Adamantyl dominates. 4/20 top hits use adamantyl tail (positions #1, #5, #8, #13, #16). Three-dimensional cage bites into the pocket while reducing hERG-liable planar surface.
  2. Halogen variants unlock the scaffold. The new R3 = −Cl (#1, #3, #4, #20) and R3 = −CN (#7, #8, #9, #14) substituents, absent from v3b, populate the top ranks. v3b only had −CF3/−F/−Me/−H on 2-amino-quinoline-3.
  3. logP reduction is huge. v5 top-20 logP range [−0.47, 2.88], median 1.51. v3b top-1 was 5.47. This is a >3 log-unit reduction → dramatic hERG class-liability reduction per STRC h01 Phase 6c hERG Extension 2026-04-24 pharmacophore analysis.
  4. Vina and APBS reward different chemistry:
    • Best Vina-only: 1-naphthyl__CONHOH__-Cl (−7.04) — big planar tail, weak pocket Coulomb (−0.41)
    • Best APBS-only: 1-naphthyl__CONHOH__-F (−1.86) — same scaffold, halogen-swap boosts Coulomb via C-F dipole bias
    • Best combined: adamantyl__CONHOH__-Cl balances both — 3D tail + favorable Coulomb
  5. Combined-score winner reflects NEW chemistry. v5 top-1 is not a tweak of v3b top-1 — it’s a distinct scaffold class (adamantyl replacing naphthyl, Cl replacing CF3). This shows the Coulomb-aware objective actively redirects medchem away from Vina’s local maxima.
  6. Phase 5k predicted it. Phase 5k-B top-1 was 2-amino-quinoline-3__1-naphthyl__CONHOH__-CF3 with ensemble ΔG −2.86 kcal/mol; v5 top combined doesn’t beat that raw APBS number yet (max ΔG_APBS here is −1.86 on 1-naphthyl__CONHOH__-F) but wins on combined via Vina complementarity. The next-round design direction is to grow adamantyl-class chemistry and sample the Coulomb-optimal tail/R3 jointly.

Interpretation

  • The Coulomb-aware objective fundamentally changes what gets designed. v5 top-20 has a completely different chemical character from v3b top-20 — heavier sp3, sub-3 logP, no CF3-only bias, halogen diversity. This is the forward path for Misha.
  • Vina alone cannot find these molecules. All top-20 v5 combined hits rank worse on Vina-only than v3b top-1 ensemble. A Vina-only medchem iteration would have filtered them out.
  • Every top hit has 85-100% ensemble-positive Coulomb — meaning across 20 Phase 5d mutant MD snapshots, the pocket is electrostatically welcoming in the overwhelming majority of dynamic states. This is a robustness guarantee v3b never had (v3b poses fell more in pocket entrance regions that varied more across dynamics).
  • hERG improvement is structural, not incidental. logP reduction is a side-effect of replacing hydrophobic planar tails (naphthyl/biphenyl, logP ≈ 5) with hydrophilic non-planar tails (adamantyl/BCO/cyclohex, logP ≈ 2). Combined with out-of-plane geometry, this removes both known hERG pharmacophore liabilities from STRC h01 Phase 6c hERG Extension 2026-04-24.

Limitations

  1. Single-frame Vina dock. Phase 8b docked into static E1659A only (not ensemble); v3b scored ensemble-mean across 5 snapshots. Direct numerical comparison requires re-docking v5 on the same 5 snapshots. Mitigated: all v5 ligands were scored the same way, so intra-v5 ranking is valid.
  2. α = 1.0 unit-weight. The combined score weights Vina and APBS equally in kcal/mol. This is a reasonable zeroth-order choice but assumes their error bars are comparable. Bayesian calibration against experimental Kd data (when available) would refine α; for now α = 1 is a tractable starting point.
  3. Only CONHOH heads in the top-20. The phosphonate / sulfamate / CONHOMe / tetrazole variants were generated but dropped by design-priority filter or failed OA-centroid extraction (tetrazoles use N atoms, not OA). Next iteration should extend the acid-head centroid logic to handle tetrazolide/phosphonate-O/sulfamate-O.
  4. ADMET not yet run on v5. Phase 7 ADMET-AI triage is needed before declaring any v5 hit wet-lab-ready. Queued as Phase 8c overnight.
  5. Off-target panel not re-run. Phase 6c was on v3b leads. v5 non-planar chemistry should be structurally less hERG-liable, but this needs direct re-test. Queued as Phase 8e overnight.
  6. Library still small (60). 60 ligands is a design-probe, not a full screen. Phase 8d v5.2 should expand to ~200-300 covering the full non-planar × head-bioisostere × R3 space.

Ranking delta

  • Hypothesis h01: tier A held | mech 3 held (this phase advances chemistry forward, not mechanism) | deliv 3 → 3 held but upgrade-path unlocked (v5 logP 2 vs v3b logP 5 removes systemic-exposure toxicity barrier; intracochlear requirement from Phase 6c may be revisitable after ADMET+off-target on v5) | misha_fit 4 held
  • Next-step update:
    • Phase 8c ADMET-AI triage on v5 top-20 (queued)
    • Phase 8d v5.2 library expansion (200+ ligands with full head × tail × R3 cross-product; queued)
    • Phase 8e off-target panel re-test on v5 top-5 (queued)
    • Phase 5d-WT (matched full-length WT MD, 2 ns) for Phase 5c-mutant cryptic-pocket specificity claim (queued — uses GPU, runs overnight)
    • v5 top-1 candidate for wet-lab shortlist consideration: v5__aq3__adamantyl__CONHOH__-Cl
  • mech=4 upgrade path unchanged: holo-MD + MM-GBSA, OR experimental Kd binding to E1659A pocket (off-computational track).

Connections

Graph View

Backlinks