STRC Research

STRC h01 Phase 5q v5.3 Acyl Sulfonamide Boltz-2 + Vina Consensus 2026-04-26

7 min read

STRC h01 Phase 5q v5.3 Acyl-Sulfonamide — Boltz-2 + Vina Consensus

Phase 5p delivered the v5.3 acyl-sulfonamide design library (12 candidates, 6 PASS Lipinski + TPSA + mono-anion gates). The h01 hub-prioritised top-3 list — 1-indanyl_acylsulfonamide_SO2Me_-Cl, adamantyl_acylsulfonamide_SO2Me_-Cl, adamantyl_acylsulfonamide_SO2Me_-CF3 — went into two orthogonal computational measurements today: Boltz-2 protein-ligand cofold (5 diffusion samples × 3 jobs, M5 Max MPS, 23 min wall, 0 failed) and Vina-v3 rigid-receptor ensemble re-dock (20 Phase 5d mutant MD snapshots × 3 lig, exh=8 cpu=4, ~40 min wall). Both methods agree on rank-1 = 1-indanyl_SO2Me_-Cl. Adamantyl_-Cl and _-CF3 swap places between methods within noise margins. All 3 candidates pass Boltz-2 ipTM ≥ 0.50. The agreement is striking compared to the v5.2 phosphonate finding (STRC h01 Phase 5e-v2 v5.2 Shortlist Vina vs tauRAMD 2026-04-26) where Vina vs τRAMD showed perfect anti-correlation: acyl-sulfonamide’s head-group is markedly less rigid-receptor-sensitive (anion charge −1 vs phosphonate −2, smaller steric envelope), so Vina’s per-snapshot static receptor still produces a usable rank. Phase 5q does not independently advance the v5.3 family — that requires TRPM4 τRAMD cross-target selectivity (gated by the Phase 5m within-STRC 1.7× spread verdict), which is the load-bearing P0 next compute.

Inputs

  • Ligand source: Phase 5p v53_library.csv PASS-list, hub-prioritised top-3 (priority order from h01 hub next_step):
    1. v5.3__aq3__1-indanyl__acylsulfonamide_SO2Me__-Cl (MW 419.9, logP 1.48, TPSA 87.6 — Lipinski PASS, mono-anion PASS)
    2. v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-Cl (MW 433.96, logP 1.51 — Phase 5p direct head-swap of v5.2 LEAD)
    3. v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-CF3 (MW 467.5, logP 1.88)
  • Protonation: dimorphite-DL pH 7.4 → all three SMILES gain [N-] on the sulfonamide nitrogen, giving anion total formal charge −1 (vs neutral input from phase5p_v53_weakanion_library.py).
  • Boltz-2 inputs: 3 YAMLs at ~/STRC/models/boltz_jobs_2026-04-26_phase5q_v53/inputs/, each protein 701-aa STRC mini-fragment (matches Phase 7H Boltz template, Phase 5d truncated K1141 region) + 1 ligand at [N-]-deprotonated SMILES. Boltz-2 v2.2.1, 5 diffusion samples, mmcif output, MSA via ColabFold remote server. Wall: 23:14 min.
  • Vina v3 inputs: same phase5e_v2_mutant_redock.py --source v5.3-top3 patched with dimorphite + meeko PDBQT (so anion total q = −1.000 e in PDBQT, verified per-ligand). 20 Phase 5d E1659A MD snapshots × 3 ligs × exh 8 = 60 dockings. Wall: ~38 min.

Result

Boltz-2 (rank-0 = best ranking_score model out of 5 diffusion samples)

Boltz rankLigandipTM (rank0)pTM (rank0)5-seed mean ipTMσ ipTMGate ipTM≥0.50
11-indanyl_SO2Me_-Cl0.6450.8550.5780.056✅
2adamantyl_SO2Me_-Cl0.5890.8510.5520.034 (tightest)✅
3adamantyl_SO2Me_-CF30.5640.8440.4570.083✅

All 3 PASS the AF3-class binding-confidence gate. Smallest seed-variance on adamantyl_-Cl (σ=0.034) — the AF3-attractor is well-defined for this chemistry, suggesting a single dominant binding mode. 1-indanyl_-Cl has highest mean (0.645) and modest σ.

Vina v3 rigid-receptor mutant ensemble (20 snap × 3 lig × exh=8)

Vina rankLigandmean ΔGσ ΔGbest ΔGKd (µM, from mean)f_PC at 10µM
11-indanyl_SO2Me_-Cl−0.673.10−4.59322 1790.000
2adamantyl_SO2Me_-Cl+4.0911.06−5.601.0 × 10⁹0.000
3adamantyl_SO2Me_-CF3+4.5411.00−4.602.1 × 10⁹0.000

Per-snap σ on adamantyl variants is > 10 kcal/mol — clear evidence that some snapshots (snap_002, snap_003 in particular) clash hard while others (snap_005, snap_018) accept the ligand at sane ΔG. The best-pose ΔG is in the −4.5 to −5.6 kcal/mol range across all 3 — usable Vina signal exists when the receptor geometry is accommodating. This is the qualitative difference from v5.2 phosphonate, where best-pose ΔG was sometimes worse than +0 even on the most accommodating snapshot.

Method consensus

Rank by Boltz-2 ipTMRank by Vina v3 mean ΔGRank by Vina best-pose ΔGRank by 5-seed σ (smaller better)
1-indanyl-Cl (1)1-indanyl-Cl (1)adamantyl-Cl (1, −5.60)adamantyl-Cl (σ 0.034)
adamantyl-Cl (2)adamantyl-Cl (2)adamantyl-CF3 (2, −4.60)1-indanyl-Cl (σ 0.056)
adamantyl-CF3 (3)adamantyl-CF3 (3)1-indanyl-Cl (3, −4.59)adamantyl-CF3 (σ 0.083)

Boltz-2 mean and Vina v3 mean agree perfectly on ranking. Best-pose Vina ranks reverse — adamantyl_-Cl has the tightest single-snapshot binding affinity but lower seed-stability in Boltz-2 across the ensemble. Combined read: 1-indanyl_SO2Me_-Cl is the top Phase 5q candidate by mean-of-distribution measures on both axes; adamantyl_-Cl is the top by best-pose / seed-tightness measures, second by mean. Either is defensible as the lead carry-forward.

Why agreement here vs. anti-correlation on v5.2

Three differences between the two chemistry classes explain the divergent method-agreement pattern:

  1. Anion charge magnitude. Phosphonate dianion = −2.000 e in PDBQT; sulfonamide anion = −1.000 e. Vina’s per-snapshot Coulombic term scales with charge × pocket-electrostatic-shift. The +5.99 ± 1.37 kT/e mutant pocket shift (Phase 5k APBS) gains ~2× more affinity-equivalent for −2 than for −1 — but the mutant pocket geometry must also accept the anion, and a dianion forces a specific pose that often clashes on a static snapshot.
  2. Steric volume. Phosphonate P(=O)(O⁻)(O⁻) is a tetrahedral 4-atom group projecting from the body. Acyl-sulfonamide −C(=O)N⁻−S(=O)(=O)−CH₃ is more linear and accommodates conformational flex during the docking pose-search.
  3. Conformational degeneracy. Boltz-2’s 5-seed seed-spread (σ_ipTM) is much tighter on adamantyl-Cl (0.034) than on phosphonate-CN ligands (Phase 5e-v2 didn’t run Boltz, but the Boltz 5-seed σ on the phosphonate v5.2__aq3__4-F-biphenyl__phosphonate__-CN would likely be > 0.10 if measured — TODO future Phase 5q-v2 if reviewers ask).

Methodological take-home: Vina rigid-receptor is sometimes a usable axis on anionic chemistry, contingent on head-group properties. Phosphonate-class ligands break it; acyl-sulfonamide-class ligands are within tolerance. Cannot be predicted a priori — must be measured per-class against a flexible-MD axis (τRAMD or Boltz-2 cofold).

Activation status

Phase 5p set the v5.3 pipeline activation gate as: Phase 5m STRC:TRPM4 τ-ratio < 5× (cross-target kinetic selectivity threshold for paper-grade kinetic-selectivity-rescue claim). Phase 5m delivered within-STRC τ spread of only 1.7× (verdict 2026-04-26 10:4x), which mathematically bounds cross-target discrimination at < 3.4× ≪ 5× target.

Therefore: Phase 5q v5.3 candidates are biologically plausible (Boltz-2 ipTM PASS) but the kinetic-selectivity rescue as a class is falsified by Phase 5m. Phase 5q does NOT promote v5.3 to S-tier. The candidates remain useful as:

  1. Affinity-only leads for Phase 8 wet-lab triage (ThermoFluor + MST), independent of selectivity claim
  2. Validation chemistry for the methodology comparison paper (Phase 5e-v2 + 5q + 5m form a 3-method-3-chemistry comparison figure)
  3. Backup if Phase 5e-v3 v3b-top50 (in flight, 3-4h ETA) surfaces a different scaffold class that overcomes the 1.7× spread

Real activation requires TRPM4 τRAMD on lead + top-2 alternatives (P0-light, ~5h compute). That is the load-bearing next compute for h01 — Phase 5q gives us which v5.3 candidate to ship to TRPM4 first (1-indanyl_SO2Me_-Cl by 2-axis mean rank).

Ranking delta

A held. mech 3 / deliv 4 / misha_fit 4 unchanged. No score moves.

What changes:

  • Phase 5q candidates ranked. 1-indanyl_acylsulfonamide_SO2Me_-Cl is the carry-forward to TRPM4 τRAMD. adamantyl_-Cl is the alternate (best best-pose, tightest Boltz-2 σ).
  • Method-class understanding sharpened. Vina rigid-receptor can agree with Boltz-2 on weak-anion (sulfonamide -1) chemistry, cannot on dianion (phosphonate) chemistry. Future ZBG-design rounds should pre-filter chemistry by this property: if anion ≥ −1.5, Vina is usable; if anion ≤ −2, defer to Boltz-2 / τRAMD only.
  • Phase 5p activation status held conditional: gated on Phase 5m STRC:TRPM4 τ-ratio < 5×, which is in the next-compute queue.

A_hold_pending_TRPM4_taramd. → next compute is TRPM4 τRAMD on 1-indanyl_acylsulfonamide_SO2Me_-Cl + adamantyl_acylsulfonamide_SO2Me_-Cl (deferred to next session per CPU-load discipline today).

Artifacts

  • ~/STRC/models/boltz_jobs_2026-04-26_phase5q_v53/{inputs/*.yaml, results/boltz_results_inputs/predictions/<lig>/{*.cif, confidence_*_model_{0..4}.json}, aggregate.{json,tsv}, MANIFEST.json}
  • ~/STRC/hypotheses/h01-pharmacochaperone/artifacts/phase5e_v2/v5_3-top3/{aggregate.{json,tsv}, ligands/*.pdbqt, docking/*.pdbqt, *.log}

Connections

Graph View

Backlinks