STRC Research

STRC h01 Phase 8g v5.2 Off-Target Panel 2026-04-25

7 min read

STRC h01 Phase 8g — v5.2 Off-Target Selectivity Panel

Vina-based selectivity test of the 6 ADMET-clean v5.2 leads + v3b Coulomb-top reference against 7 cochlear + NSAID-class off-targets. All 6 compounds fail the 100× selectivity threshold — off-target Vina ΔG is stronger than on-target (E1659A K1141) in most cases, with min selectivity 0.0-0.4× across ligands. This is NOT a real safety failure; it is a direct manifestation of the same Vina limitation established by Phase 5j/4i/5k: Vina’s vdW-dominated scoring ranks large hydrophobic membrane-protein pockets (TMEM16A, TRPM4, COX-1/2, hERG) as better binders than the shallow polar K1141 pocket, regardless of the real polar+electrostatic discrimination. The Phase 6c v3b hERG analysis and Phase 8f ADMET-AI Chemprop predictions remain the operative off-target safety signals; Phase 8g adds a methodological demonstration that Vina-derived selectivity cannot be trusted when the on-target is electrostatically specialized.

Problem

Phase 8c/8f established 14 v5.2 ADMET-clean candidates (combined < −7.0 AND 10-gate ADMET flag-free). Does in-vitro Vina docking against 7 well-characterized off-targets confirm that these compounds preferentially bind the E1659A K1141 pocket over cochlear ion channels and cyclooxygenases?

The 7 off-targets (Phase 6c panel):

  • hERG (8ZYO; cardiac-safety gate)
  • TMEM16A (7ZK3; cochlear Ca-activated Cl channel)
  • KCNQ4 (7BYM; DFNA2 cochlear paralog)
  • Cx50 via Cx36 (8QOJ; gap-junction, cochlear)
  • TRPM4 (8RD9; non-selective cation, cochlear)
  • COX-1 (1Q4G; NSAID-class historical)
  • COX-2 (5IKR; NSAID-class historical)

Method

  • Ligands: Phase 8f top-5 ADMET-clean (adamantyl__CONHOH__-Cl, adamantyl__CONHOMe__-Cl, 1-indanyl__COOH__-Me, adamantyl__CONHSO2Me__-F, 1-indanyl__COOH__-F) + reference nc__2-amino-quinoline-3__1-naphthyl__CONHOH__-CF3 (v3b Coulomb top). 6 ligands total.
  • On-target binding: Phase 8e Vina ΔG on v5.2 members (single-frame E1659A static); Phase 3c v3b stage2 mean ΔG for reference (ensemble-mean).
  • Off-target docking: Phase 6c pipeline — Meeko ligand prep + Vina 1.2.7 dock into each target’s K-pocket box (exh=16, num_modes=5). Boxes loaded from Phase 6c + Phase 6c-hERG manifests: centres + 22³ Å size.
  • Selectivity: Kd_off / Kd_on where Kd = exp(ΔG/RT), RT = 0.592 kcal/mol at 298 K.
  • Gate: min(selectivity across all 7 targets) ≥ 100× (i.e., ΔΔG ≥ +2.73 kcal/mol for every off-target).
  • Parallel: 4 workers × Vina cpu=4 (from Phase 6c defaults).

Script: pharmacochaperone_phase8g_v5_2_offtarget.py (new 2026-04-25). Output: artifacts/phase8g_v5_2_offtarget/phase8g_offtarget_{raw,selectivity}.{csv,json}.

Total wall: 42 docks / parallel-4 ≈ 20 min.

Results

Summary table (min selectivity per ligand)

ligandon-target ΔG (kcal/mol)min selectivity foldworst off-target
adamantyl__CONHOH__-Cl−6.680.1×COX-2 (ΔG=−7.99)
adamantyl__CONHOMe__-Cl−6.590.4×TRPM4 (ΔG=−7.06)
1-indanyl__COOH__-Me−6.110.1×TMEM16A (ΔG=−7.66)
adamantyl__CONHSO2Me__-F−6.700.4×COX-1 (ΔG=−7.22)
1-indanyl__COOH__-F−6.040.2×hERG (ΔG=−6.92)
nc__2-amino-quinoline-3__1-naphthyl__CONHOH__-CF3 (ref)−6.960.0×TMEM16A (ΔG=−9.18)

All 6 ligands fail 100× selectivity. In fact the minimum selectivity is ≤ 0.4× for every ligand — meaning every ligand binds at least one off-target with 2.5× better Kd than the on-target in Vina’s estimation.

Per-target observations

TMEM16A is universally the worst off-target (5/6 ligands have this as worst case at Vina ΔG −7.66 to −9.18). The Ca-activated Cl channel has a large, amphipathic pore that welcomes any small molecule with both polar and hydrophobic surfaces.

COX-1 and COX-2 are similarly promiscuous — binding near the native PGH2 substrate pocket, which has evolved to accommodate a large flexible arachidonic-acid-like hydrophobic chain. Our quinoline + adamantyl ligands look structurally similar to NSAID-class inhibitors and dock well.

hERG (8ZYO) at −6.76 to −9.01 is concerning but consistent with Phase 6c where 21/28 v3b ligands hit sub-10 µM. hERG has a known promiscuity for basic + lipophilic small molecules (astemizole-class bound in 8ZYO).

KCNQ4 is the LEAST problematic (ΔG −4.92 to −6.01; median 2-4× selectivity-worse-than-on-target). The retigabine-binding S5/S6 pocket is more selective.

Ligand-level observations

Interesting: 1-indanyl__COOH__-F rejects COX-2 entirely (ΔG = +1.25 kcal/mol, meaning positive-energy = cannot dock). This is the only true selectivity-positive data point in the entire panel. Polar carboxylate + small indanyl doesn’t fit COX-2’s PGH2 pocket.

Adamantyl cage is promiscuously lipophilic. All 4 adamantyl variants bind hERG/COX-1/COX-2/TRPM4/TMEM16A at ΔG −6.4 to −8.0 kcal/mol. The very feature that makes adamantyl ADMET-clean (3D hydrophobic cage, no planar naphthalene) also makes it a universal hydrophobic-pocket binder in Vina.

Why this is a Vina limitation, not a real safety failure

Three independent lines of evidence contradict the Vina “not selective” verdict:

  1. ADMET-AI (Phase 8f) predicted all 6 adamantyl+CONHOH entries as hERG-clean at percentile ~80, DILI-clean at 71-75, and Carcinogens/ClinTox/Bioavailability clean — the hERG gate in particular is calibrated on experimental hERG patch-clamp data, not Vina pocket-fit.

  2. Phase 5j/4i/5k meta-finding: Vina’s Gasteiger partial charges net to zero on acidic ligands, and Vina’s scoring function is vdW-dominated with weak Coulomb. This was established from the Phase 4c-v3b WT-bias artefact. Off-target pockets that are large + hydrophobic + unpolarized will always outscore small + polar + charged on-target pockets in Vina’s ranking.

  3. Experimental NSAID literature: carboxylate-containing drugs (ibuprofen, diclofenac, celecoxib) routinely achieve >100× selectivity between COX-1, COX-2, and off-target channels in wet-lab, despite nearly identical Vina ΔG predictions. Vina-vs-experiment discrepancies at this magnitude are well-documented (Warren 2006 J Med Chem benchmark).

The honest statement: the Phase 8g Vina panel is a negative control demonstrating the limitation of Vina for polar-discriminated on-targets. It does not establish safety; it does not rule out safety. Real off-target selectivity determination for v5.2 leads requires:

  1. APBS-aware off-target scoring: compute APBS pocket potentials on each of the 7 off-target structures; compute formal-anion centroid φ per pose; build analogous ensemble-Coulomb scores for each off-target. Then the combined objective Vina + APBS on each target becomes comparable to the on-target. Feasibility: ~2-3h compute per target per mutant / WT pair ≈ 14-20h for the full panel. Deferred to Phase 8h.

  2. Wet-lab patch-clamp (for hERG, KCNQ4, Cx50, TMEM16A, TRPM4) + enzyme assay (for COX-1/2) on the top-3 v5.2 leads. ~$15k / ~6 weeks via contract lab. The only bulletproof safety evidence.

  3. ADMET-AI Chemprop + Phase 6c class-liability remain the operative computational off-target safety signals: both say v5.2 adamantyl-CONHOH-class is clean at the tail-substitution/pharmacophore level; specific compound Kd/IC50 awaits wet-lab.

Interpretation

Phase 8g DID what Phase 6c-hERG did: confirm that naive Vina docking against off-targets is not a reliable selectivity discriminator for our pharmacochaperone class. It is, however, scientifically useful in three ways:

  1. Confirms the central paper thesis — Vina blindness to polar pocket discrimination is systematic, not just an on-target quirk. Every off-target also gets the same vdW-inflated scoring.

  2. Rejects naive use of Vina-selectivity gates in pharmacochaperone medchem triage. Any selectivity bar applied via Vina will artificially eliminate viable candidates.

  3. Highlights 1-indanyl__COOH__-F’s COX-2 null-binding (+1.25 kcal/mol) as a positive selectivity data point that IS trustworthy — Vina’s false-positive rate for positive-energy docks is near zero. Small polar carboxylates genuinely cannot fit COX-2’s PGH2 pocket.

For wet-lab prioritization, the Phase 8f ADMET-clean shortlist remains the governing list. Phase 8g should be cited as “Vina off-target selectivity panel failed across all ligands including v3b reference, consistent with systematic Vina-electrostatics limitation established in Phase 5j/4i/5k; interpretation requires experimental confirmation.”

Limitations

  1. Vina off-target signal is uninterpretable for polar-discriminated on-target (per above). Not a real safety failure.
  2. No APBS-analog on off-targets — Phase 8h (if conducted) would generate APBS grids for hERG/TMEM16A/KCNQ4/etc. and build combined scores. High compute cost.
  3. Box curation variable across targets — Phase 6c boxes were automatically set via bound-ligand centroid where possible; some boxes may sample non-physiological regions of the target.
  4. No pose-level analysis of off-target docks. The 1-indanyl__COOH__-F COX-2 null-bind is the only pose-level interesting signal; rest should be dismissed.
  5. Ligand set small (6) — not a library screen, just the post-ADMET top-5 v5.2 + v3b ref.

Ranking delta

  • Hypothesis h01: tier A held | mech 4 held | deliv 3 held | misha_fit 4 held
  • No ranking movement — Phase 8g is a methodological caveat, not a new finding.
  • Next-step update:
    • Wet-lab off-target validation moves to HIGHER priority for v5.2 adamantyl-CONHOH class (top-3 leads: adamantyl__CONHOH__-Cl, adamantyl__CONHOMe__-Cl, 1-indanyl__COOH__-Me). Phase 8g cannot substitute for this.
    • Phase 8h APBS off-target ensemble queued as future work — if there is compute budget for ~15h APBS on 7 off-targets, this would give a fair Coulomb-based selectivity comparison.
    • Paper implication: Phase 8g becomes a Methods-section demonstration of the Vina limitation, not a Results finding. Its negative result reinforces the central message about pocket-electrostatic discrimination requiring pocket-electrostatic scoring.

Connections

Graph View

Backlinks