STRC Research

STRC h01 Phase 8h-lite · Light Computational Evidence Package 2026-04-26

5 min read

STRC h01 Phase 8h-lite · Light Computational Evidence Package

Date. 2026-04-26 night. Goal. Close five paper-claim gaps for H01 with light compute (<1 min total wall time, no new MD/dock/APBS), using existing Phase 5d/5k/8e data + RAG-mined literature (Salt 2001, Halgren 2009 SiteMap, Schacht 2008 Auditory Trauma). Compute. 6 Python scripts at hypotheses/h01-pharmacochaperone/scripts/phase8h_lite/. Total run time <30 s on M5 Max.

Five proofs

#1 K1141 pocket druggability (Halgren SiteMap-style)

druggability_score.py — Python grid burial estimator (fpocket replacement after Homebrew binary qhull bug). Halgren 2009 J Chem Inf Model 49:377 spec.

DescriptorValueDruggable benchmarkPass
Volume V1145 ų250–1000 (avg 280, range to 1500)✅ large pocket
Phobic enclosure0.609≥ 0.40 (transthyretin tafamidis 0.55)✅
Philic enclosure0.391≤ 0.50✅
Phobic / philic1.56> 1.0✅
V_pocket / V_lig2.751.5–3.0 (canonical)✅

Lining: aromatic-rich phobic shell (W1612, F1646, F1169, W1652) — π-stacking surface for adamantyl/aromatic core. K1141 lines 26% of philic shell — the targeted formal-anion docking site.

Verdict: ✅ K1141 pocket meets all 4 classical druggability gates. Phase 5c GREEN now has a quantitative number.

#2 RWM permeability prediction

rwm_permeability.py — Salt 2001 P_TMPA baseline + Stokes-Einstein MW^(−1/3) + Avdeef logP/TPSA + Henderson-Hasselbalch f_neutral.

QuantityLeadTMPA (Salt 2001)
P (eff @ pH 7.4)1.45×10⁻⁷ cm/s1.9×10⁻⁸ cm/s
Relative to TMPA7.6×1×
Inferred basal-turn concentration @ 90 min~60% applied~8% applied

Sensitivity: CONHOH pKa 8.5 vs 9.0 → P ratio 7.2× vs 7.6× — robust.

Verdict: ✅ Lead is plausibly deliverable as a topical RWM ear-drop. logP 1.94 more than compensates the ~24% Stokes-Einstein MW penalty.

#3 Off-target electrostatic selectivity (INCONCLUSIVE LITE)

Two attempts: selectivity_charge_proxy.py (Vina-pose-anchored) + selectivity_pocket_scan.py (receptor-wide K/R clusters).

Both proxies fail to discriminate properly:

  • Vina parked the lead’s CONHO⁻ at 7.5 Å from K1141 (the same Gasteiger-neutral-charge artefact Phase 5k diagnosed) — pose-anchored count gives only +1 net for E1659A.
  • Receptor-wide scan returns large K/R clusters on all 4 off-target channels (TRPM4 +10, KCNQ4 +8, TMEM16A +7, Cx50 +5) because their voltage-sensors / selectivity filters are K/R-rich. These are NOT enclosed druggable pockets — the count doesn’t capture what Phase 5k measured.

Verdict: ⚠ INCONCLUSIVE. Proper selectivity test requires APBS on enclosed pockets per off-target. This is the heavy planned step. Salvage: Vina off-target binding indistinguishable from on-target binding (Phase 8g), but Phase 5k’s mechanism is Coulomb-driven on a single buried Lys in a small enclosed hydrophobic pocket — none of the 4 off-target structures presents an obvious such site by inspection. Wet-lab off-target panel remains required.

#4 Pose ensemble stability (LIE substitute)

pose_ensemble_stability.py — distance K1141 NZ ↔ lead CONHO⁻ across all 20 Phase 5d snapshots.

MetricValue
K1141 NZ position drift mean2.48 Å
K1141 NZ position SD1.11 Å
K1141 NZ per-axis SD (x/y/z)1.72 / 1.15 / 1.10
Lead CONHO⁻ ↔ K1141 NZ mean5.02 Å
Lead CONHO⁻ ↔ K1141 NZ SD0.57 Å
Range3.93–6.18 Å

Verdict: ✅ Pocket is rigid (SD ~1 Å per axis) and lead Vina pose sits in Coulomb-attraction range across the entire ensemble. Phase 5k +5.99 kT/e isn’t a single-frame artefact. Vina pose at 5 Å is on the edge of direct salt-bridge zone (canonical 2.7–3.0 Å); APBS-correct mode optimization would tighten to ~3 Å, contributing additional ~−3 kcal/mol on top of Phase 5b’s −8.19 kcal/mol — projected Kd ~5–10 nM (incremental hypothesis to wet-lab, not measurement).

#5 Tanimoto vs known cochlear ototoxins

tanimoto_ototoxins.py — RDKit Morgan FP r=2 2048 bits, lead vs 12-compound class panel from Schacht 2008 Auditory Trauma, Protection and Repair (RAG-indexed).

Class coverage: aminoglycosides (gentamicin/kanamycin/amikacin/neomycin), loop diuretics (furosemide/ethacrynic), platinum (cisplatin/carboplatin), salicylates (aspirin), macrolide (erythromycin), quinoline (quinine), glycopeptide (vancomycin).

Closest matchTanimoto
aspirin0.127
furosemide0.120
aspirin (next)0.103

Threshold for flagging: 0.40. Max 0.127 — well below.

Verdict: ✅ Lead has NO Morgan-FP motif overlap with any major ototoxin class. Note this is shared-substructure only; mechanism-specific transporter cross-reactivity (cisplatin via CTR1, aminoglycosides via TRPML/MET) is not predicted from chemical similarity — wet-lab cytotox remains required.

What changes for the paper

Four new claim-cells unlocked with quantitative numbers (vs adjectives):

  • Druggability of K1141 pocket — table values for V, phobic, philic, V_pocket/V_lig
  • Predicted topical-delivery PK with sensitivity envelope
  • Mech-4 ensemble validation on actual lead geometry (not generic anion probe)
  • Chemical-class clean tox baseline

One claim still requires the heavy step: off-target selectivity (APBS on enclosed pockets, 5 receptors).

Ranking delta

  • tier A held
  • mech 4 held (Phase 5k still load-bearing; pose-stability #4 reinforces but doesn’t elevate)
  • deliv 3 → 4 (RWM permeability prediction closes the “topical ear-drop” delivery claim with a quantitative number)
  • misha_fit 4 held
  • next_step: APBS on off-target enclosed pockets is now the highest-priority heavy step (#3 inconclusive lite); paper draft prose can be filled in for sections 1–2, 4–5 immediately

Connections

Graph View

Backlinks