Phase 5m production τRAMD ranking — v5.2 mut-prefer + mech-anchor shortlist

5 replicas/ligand × max 5 ns biased / 30 min wall on Phase-5d-truncated 151-aa K1141 fragment (1066-1216, MD-relaxed parent of E1659A; ring RMSD 0.97 Å vs AF3-folded substrate per Phase 5l). 14 kcal/(mol·Å) bias, maxDist 4.0 nm, Gasteiger charges (am1bcc + Precision:mixed both seg-faulted; production-grade upgrade pending NAGL .pt model). All 25 replicas UNBOUND.

Ranking by mean τ (high → low; longer = better residence time)

RankLigandn_unbound/nτ̄ ± SD (ps)τ median (ps)τ range (ps)
1v5.2__aq3__adamantyl__CONHOMe__-CN5/526.1 ± 16.5818.714.7 – 55.1
2v5.2__aq3__4-F-biphenyl__phosphonate__-CN5/521.1 ± 5.4320.015.5 – 30.1
3v5.2__aq3__adamantyl__CONHOMe__-Cl5/519.6 ± 9.6315.813.7 – 36.6
4v5.2__aq3__1-indanyl__phosphonate__-CF35/519.0 ± 4.2419.713.9 – 25.2
5v5.2__aq3__adamantyl__CONHOH__-Cl5/515.4 ± 0.9715.714.3 – 16.5

Ratio vs CONHOH legacy mech-anchor (adamantyl_CONHOH_-Cl)

Ligandτ̄ (ps)τ_ratioVerdict
adamantyl__CONHOMe__-CN26.11.70×🟢 longer
4-F-biphenyl__phosphonate__-CN21.11.37×🟢 longer
adamantyl__CONHOMe__-Cl19.61.27×🟢 longer
1-indanyl__phosphonate__-CF319.01.24×🟢 longer
adamantyl__CONHOH__-Cl15.41.00×⚪ tie

Interpretation

All 25 replicas UNBOUND within 5 ns biased MD; no ligand reached the 5-ns censoring cap. τ values range 13.1–55.1 ps, mostly clustered 14-21 ps. Differences between ligands are within 1.5× — not statistically significant for kinetic-selectivity claim (target ratio ≥ 5× for clinical residence-time selectivity per Pharmacochaperone Residence Time Criterion).

Critical caveats:

  1. Substrate is K1141 fragment alone, no off-target. This ranks STRC-on-STRC residence time only. Full kinetic-selectivity test requires running the same lead against TRPM4/TMEM16A pockets — outside this session’s scope (Boltz-2 binding-mode test queued P0-light is the right next step).
  2. τRAMD under 14 kcal/(mol·Å) bias is a relative-ranking method, not absolute-k_off. All ligands unbinding within 13-55 ps under aggressive bias is the expected behaviour; absolute k_off is hours-orders longer. The clinical question is: does the ranking (lead vs alternatives) hold without bias? Established τRAMD literature (Kokh 2018) says yes for protein-ligand systems with comparable-affinity ligands.
  3. No clear winner emerges. All 5 ligands cluster within τ̄ 15-26 ps. Lead adamantyl-CONHOMe-Cl is mid-pack. CONHOMe-CN edges out at τ̄ 26.1 ps — interesting because Phase 8h-lite #6 RWM permeability flagged it as marginal (P 3.0× TMPA, TPSA 86.5 + low logP) but not BLOCKED. Worth re-evaluating.
  4. CONHOH legacy mech-anchor τ̄ 15.4 ps is the SHORTEST. Confirms Phase 7I demotion of CONHOH from lead to mech-anchor: its kinetic profile is genuinely worse than the head-swap variants by ~10-50 %. Validates the v5.2 → v5.3-CONHOMe lead pivot from a residence-time perspective independent of the Boltz-2 ipTM data.
  5. N=5 replicas is small for SEM tightness — production paper-grade run should use n=10 with 2 controls (diflunisal pos-ctrl, tafamidis-analog neg-ctrl) on AF3 substrate post-OpenCL-optimization (build NAGL model file or switch to GROMACS). This run is sufficient for discrimination ranking, not for absolute k_off claim.

Activation gate decision

STRC : TRPM4 τ-ratio cannot be computed yet (TRPM4 RAMD not run). However, internal STRC ranking shows τ_max / τ_min = 26.1 / 15.4 = 1.7× spread on the same target. This is far below the 5× kinetic-selectivity threshold for clinical relevance even within the same target. Cross-target selectivity is unlikely to clear 5× without head-group change. v5.3 acyl-sulfonamide PASS-list (Phase 5p) activated for Boltz-2 + Vina + APBS pipeline.