STRC Research

STRC h01 Phase 8g-v2 APBS Off-Target Rescore 2026-04-26

5 min read

STRC h01 Phase 8g-v2 — APBS Off-Target Rescore 2026-04-26

Replaces failed Phase 8g Vina off-target panel + inconclusive Phase 8h-lite #5 Coulomb proxy. Motivated by Hydroxamic Acid Divalent Cation Liability and the heavy-queue (2a) item from STRC h01 DR3 distillation log entry 2026-04-26.

Method

For each of 5 receptors (STRC E1659A K1141 reference + 4 cochlear off-targets), with crystallographic metals retained:

  1. pdb2pqr --ff=PARSE (preserves HETATM Zn/Ca/Mg/Fe with PARSE ion charges).
  2. APBS mg-auto npbe, fine grid 65³, fine box 28 Å on lead-pose centroid, coarse box ≥ protein-extent + 40 Å, εp=2.0 / εs=78.54 / 0.15 M NaCl / T=310 K.
  3. Sample φ at each hydroxamate-O atom of the Phase 8g best-pose (v5.2__aq3__adamantyl__CONHOH__-Cl MODEL 1).
  4. Sample φ in 6×6×6 Å pocket-local box around lead centroid.
  5. Record metal proximity to anion-O.

Script: ~/STRC/hypotheses/h01-pharmacochaperone/scripts/phase8g_v2/apbs_offtarget_rescore.py. Output: artifacts/phase8g_v2_apbs_offtarget.json. Wall: ~5 min, 3-way parallel.

Result

Mean φ at the lead’s hydroxamate-O atoms (high → low):

Receptoranion-O ⟨φ⟩ (kT/e)metalsmin metal–anion dist
TRPM4 (8RD9)+9.360—
TMEM16A (7ZK3)+4.896 Ca²⁺17.5 Å
STRC E1659A K1141+2.510—
KCNQ4 (7BYM)+0.910—
Cx50 / Cx36 (8QOJ)−3.020—

Margin (STRC − max-off-target) = −6.85 kT/e (STRC is 6.85 kT/e LESS attractive than TRPM4 at the lead’s anion site). Gate verdict: FAIL.

Two surprises that rewrite the story

(1) Metals are not the failure mode

DR3 §I.6 predicted Zn²⁺/Ca²⁺ chelation as the hydroxamate liability. But TMEM16A’s 6 crystallographic Ca²⁺ sit 17.5 to 62.7 Å from the lead’s anion-O — far outside coordination range (~2.4 Å). The +4.89 kT/e at TMEM16A is residue-driven, not metal-driven. TRPM4 has zero crystallographic metals and the highest φ.

Implication: bioisosteric escape via tetrazole or CONHSO₂Me (hypothesised in Hydroxamic Acid Divalent Cation Liability) may not save selectivity — any anionic warhead will be pulled into a positive pocket.

(2) STRC reference is sub-optimal in the docked pose

Phase 5k MD-ensemble-averaged K1141 pocket centre: +5.99 ± 1.37 kT/e. Phase 8g-v2 lead anion-O on STRC: +2.51 kT/e. The lead’s hydroxamate-O is sitting at less than half the maximum-φ point of its own pocket. Either (a) the Phase 8e Vina pose for STRC is sub-optimally placed for electrostatic complementarity, or (b) the K1141 maximum-φ region is not where the lead’s anion lands geometrically.

If (a): re-rank Phase 8e poses by APBS φ at anion-O instead of Vina ΔG. Expect rescue. If (b): the lead is mechanistically under-optimised for the K1141 pocket; v5.3 design needs anion-position-aware geometry, not just Coulomb-aware composition.

Caveat — Phase 8g pose quality

The off-target poses are Vina blind-dock results from Phase 8g, which itself was flagged as Vina-failure. Vina without site definition often parks ligands in the most-electrostatically-favourable surface site, not the biologically-relevant binding pocket. The +9.36 kT/e on TRPM4 may reflect a non-functional surface basin, not the actual TRPM4 ligand-binding site.

This means the FAIL verdict is conditional: “the lead’s anion is electrostatically attracted to alternative cochlear-channel surfaces with stronger Coulomb pull than its target K1141 pocket as currently posed.” That is bad — but the strength of “bad” depends on whether the off-target pockets we found are real binding sites or surface artefacts.

What this changes

Mech 4 holds. The K1141-pocket binding mechanism is intact; this proof says nothing against it. What it says is that other proteins also have positive pockets attractive to this lead.

Deliv 4 → 3. Selectivity is no longer “wet-lab to determine” — it is “computationally suspect” until corrected. A pediatric ear drop whose lead has stronger Coulomb attraction to TRPM4 than to its target cannot claim selectivity for paper §3.

Tier A held, but on warning. If the next two compute jobs (Phase 8g-v3 site-defined off-target re-dock, Phase 8e-v2 STRC pose-anion-O re-rank) do not reverse this, tier drops.

Ranking delta

  • tier: A (held, on warning)
  • mech: 4 → 4 (held)
  • deliv: 4 → 3 (selectivity gap quantified — first FAIL verdict, not just inconclusive)
  • misha_fit: 4 → 4 (held — Misha’s specific need for the lead is unchanged)
  • next_step: heavy queue reordered:
    • (NEW) Phase 8g-v3 — site-defined off-target re-dock. Use UniProt/PDB-annotated functional pockets (TRPM4 cytoplasmic Ca²⁺/CaM site, TMEM16A inner-pore lipid-binding site, KCNQ4 retigabine pocket, Cx50 hemichannel pore). Re-dock lead at each defined site, then re-run Phase 8g-v2 APBS on those poses. Expected: drop the Vina-blind-dock surface artefact contribution.
    • (NEW) Phase 8e-v2 — APBS re-rank of v5.2 STRC poses. Score all 14 ADMET-clean v5.2 poses on STRC by anion-O φ at the K1141 pocket; see whether any pose has anion-O above +5 kT/e. If yes → that lead becomes new mech-anchor; if no → v5.3 design must add anion-position-aware geometry constraint.
    • (HELD) Phase 5l fragment ladder — substrate for τRAMD; AF3 jobs scaffolded at ~/STRC/models/af3_jobs_2026-04-26_phase5l/, awaiting Egor’s AlphaFold Server upload.
    • (HELD) Phase 5m τRAMD — skeleton at scripts/phase5m_taramd_skeleton.py; depends on Phase 5l winner + ramd-openmm install.
    • (DEMOTED) v5.3 design (tetrazole/quinoxaline) — defer until Phase 8g-v3 confirms whether anion-warhead choice can rescue selectivity, or whether the issue is geometric (different question).

Connections

Graph View

Backlinks