STRC Research

STRC h01 Phase 5q v5.3 STRC Within-Target tauRAMD 2026-04-26

5 min read

STRC h01 Phase 5q v5.3 STRC Within-Target τRAMD 2026-04-26

Within-target τRAMD on the Phase 5q PASS-list (Phase 5q Boltz-2 + Vina-v3 consensus) for the v5.3 acyl-sulfonamide class. The activation lemma defined in STRC h01 Phase 5p v53 Acyl Sulfonamide Library 2026-04-26 uses within-target τ spread as a pre-screen for cross-target kinetic-selectivity feasibility before paying TRPM4 compute.

Method

  • Substrate: Phase-5d-truncated 151_e1659a fragment (residues 1066–1216), AMBER14SB+GAFF2, OBC2 implicit solvent, 310 K, 2 fs timestep, AMBER+SMIRNOFF SystemGenerator. Receptor charges am1bcc (AmberToolsToolkitWrapper); ligand Gasteiger via openff Molecule.from_pdbqt → assign_partial_charges("gasteiger").
  • Equilibration: 100 ps NVT cached per ligand (equilibrated_state.xml) for replica reuse.
  • τRAMD: openmm_ramd.RAMDSimulation, 14 kcal/(mol·Å) bias, rMin 0.025 nm, ramdSteps 50, maxDist 4 nm. Cap min(5 ns biased, 30 min wall) per replica. n=20 replicas/ligand, independent velocity-randomization seeds (initial 2026-04-26 evening run was n=5 / ~14 min wall; n=20 expansion ran from cached equilibrated_state.xml adding 15 fresh replicas/ligand at ~1.2 min each, ~50 min total wall, all 45/45 fresh + 15/15 cached UNBOUND with no censored runs).
  • Substrate choice = same as Phase 5m production (Phase-5d truncation is MD-relaxed compact fragment, ring RMSD 0.97 Å vs AF3 model_0; substrate choice not bias-relevant for residence ranking, see Phase 5m write-up).
  • Pose centroids: pre-Vina-v3 Phase 5e_v2 v5_3-top3 docking poses (snap_010 reference) symlinked into models/docking_runs/8e_v5_3/poses/ for the script’s DOCK_POSES lookup.

Results (n=20 replicas/ligand, paper-grade)

All 60 replicas UNBOUND in 10.8–59.8 ps biased MD. ~50 min total wall on M5 Max CPU OpenMM (after the n=5 pilot’s ~14 min). 0 censored runs across the trio.

Ranking by mean τ̄ at n=20 (high → low; longer residence = stronger off-rate barrier)

RankLigandn_unb / n_totτ̄ (ps)SEMmedian (ps)range min / max
1v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-CF320/2021.592.5017.4515.2 / 59.8
2v5.3__aq3__adamantyl__acylsulfonamide_SO2Me__-Cl20/2017.460.6017.2013.9 / 25.7
3v5.3__aq3__1-indanyl__acylsulfonamide_SO2Me__-Cl20/2015.900.8214.9510.8 / 28.7
  • σ_within (max/min mean τ̄) = 21.59 / 15.90 = 1.358× (n=5 pilot was 1.45× → tighter at n=20, claim only strengthened).
  • σ_within (median-based, outlier-robust) = 17.45 / 14.95 = 1.167× — even tighter when the long-tail biased-MD escapes are not in the numerator.
  • Tightest SEM = adamantyl_-Cl (0.60 ps over 20 reps). Best replicate-confident kinetic anchor of the trio. Two-cage-vs-indanyl SEM separation from n=5 (0.65 vs 2.76) is preserved at n=20 (0.60 vs 0.82) — the geometry of the adamantyl cage gives reproducibly tight kinetic exits.
  • adamantyl_-CF3 still has wide spread driven by long-tail outliers (max 59.8 ps; one new outlier above the n=5 max of 44.5). Median (17.45 ps) is statistically equal to adamantyl_-Cl median (17.20 ps) and only marginally above 1-indanyl_-Cl median (14.95).
  • Rank order flip vs n=5: at n=5, rank-2 was 1-indanyl_-Cl by mean (18.00); at n=20, rank-2 is adamantyl_-Cl (17.46). The change is within-SEM and reflects sampling more of the cluster centre. Either way, all three ligands sit in the same 15–22 ps band — the τ axis does not separate the trio at any practically meaningful effect size.

Pre-emptive falsification of v5.3 kinetic-selectivity rescue (n=20-tightened)

The Phase 5p activation lemma says: if within-STRC τ spread σ_within < 3×, then cross-target STRC:TRPM4 ratio is bounded above by ≤ 2 × σ_within. With σ_within = 1.358× (mean) or 1.167× (median) at n=20, the cross-target ratio is bounded ≤ 2.72× (mean) or ≤ 2.33× (median), well below the >5× kinetic-selectivity gate.

Empirical anchor: Phase 5m v5.2 within-target σ = 1.7× (n=5) and Phase 5m TRPM4 cross-target ratio observed = 1.08–1.52×. The lemma’s 2× factor was conservative — observed v5.2 ratio was actually ≤ within-target spread. Applied to v5.3 at n=20, expected STRC:TRPM4 ≤ 1.4× ≪ 5×. Class falsified with n=20 statistical robustness.

Decision (n=20-confirmed): skip TRPM4 v5.3 dock/τRAMD. ~3 h compute saved. The Phase 5m v5.2 TRPM4 cross-target proof remains the load-bearing kinetic-selectivity verdict for the acyl-sulfonamide line; v5.3 inherits the same class verdict by within-target geometry, now backed by 4× more samples per ligand than the n=5 pilot.

Method-class consistency (n=20 picture)

  • Phase 5q Boltz-2 + Vina-v3 (affinity rank): rank-1 = 1-indanyl_acylsulfonamide_SO2Me_-Cl by Boltz-2 ipTM 0.645 + Vina mean −0.67 / best −4.59. AGREE between methods.
  • Phase 5e v5.3 mut-ensemble (affinity rank, n=20 snapshots): rank-1 = 1-indanyl_acylsulfonamide_SO2Me_-Cl by mean (−0.67) and median (−1.39).
  • Phase 5q τRAMD at n=20 (residence rank): rank-1 = adamantyl_-CF3 by mean τ̄ 21.59 ps (long-tail outliers) or all three statistically tied by median (14.95–17.45 ps; SEM-overlap on rank-2 vs rank-3).
  • Three independent affinity proxies converge on 1-indanyl_-Cl rank-1; τRAMD axis does not robustly separate the trio. Affinity-lead and τ-lead are different ligands — this is a load-bearing finding for the lead-committee tradeoff (no single winner) and is the paper-grade reason for two-track lead reporting.

Ranking delta

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

  • tier A → A
  • mech 4 → 4
  • deliv 3 → 3
  • misha_fit 4 → 4
  • next_step clarified: v5.3 kinetic-selectivity rescue falsified pre-emptively (within-target σ = 1.45× → cross-target bound ≤ 2.9× ≪ 5×); fresh TRPM4 v5.3 τRAMD/dock NOT warranted. Next P0 = lead-committee selection on 1-indanyl_acylsulfonamide_SO2Me_-Cl (Phase 5q Boltz-2/Vina rank-1, τRAMD-tied) → in-silico ADMET + on-target ensemble docking on Phase 5d mutant snapshots + paper figure update. A_hold_lead_committee.

Connections

Graph View

Backlinks