Phase 8h-lite #3 · Off-target electrostatic-selectivity proxy (light) — INCONCLUSIVE
Goal. Light proxy for off-target selectivity: rank receptors by accessible electropositive pocket capacity, comparable to E1659A K1141 +5.99 kT/e (Phase 5k).
Method.
- Vina-pose-anchored charge proxy: count net side-chain charge within 6 Å + 10 Å of docked v5.2 lead pose (
selectivity_charge_proxy.py). - Receptor-wide pocket scan: for every Lys/Arg, count net (Lys/Arg − Asp/Glu) within 10 Å (
selectivity_pocket_scan.py).
Result
Approach 1 (Vina-pose-anchored, 6 Å + 10 Å)
| target | tight Δ | wide Δ | tight residues |
|---|---|---|---|
| E1659A STRC K1141 | +1 | +0 | LYS1173 |
| TRPM4 | 0 | +1 | none |
| KCNQ4 | 0 | 0 | none |
| TMEM16A | 0 | +1 | none |
| Cx50/Cx36 | −1 | 0 | GLU25 |
Issue. Phase 8e Vina parked the lead’s CONHO⁻ at 7.5 Å from K1141 NZ, away from the Coulomb sink. This is the same Gasteiger-neutral-charge artifact that Phase 5k diagnosed and fixed by switching to APBS. The proxy is therefore measuring “where Vina parked it” not “where the formal anion would equilibrate,” so the +1 figure for E1659A is a Vina-pose lower bound — the APBS-correct binding mode at K1141 NZ would give +3 to +4 net.
Approach 2 (receptor-wide K/R pocket scan)
| target | best K/R pocket | +K/R | −D/E | net Δ+ |
|---|---|---|---|---|
| E1659A STRC | ARG439 / LYS1104 | 7 | 2 | +5 |
| TRPM4 | ARG542 (4× Arg cluster) | 10 | 0 | +10 |
| KCNQ4 | ARG88 (Arg cluster) | 8 | 0 | +8 |
| TMEM16A | LYS273 / ARG276 | 7 | 0 | +7 |
| Cx50/Cx36 | LYS170 | 5 | 0 | +5 |
Issue. All four off-target receptors are ion channels with K/R-rich functional regions (voltage sensors, selectivity filters, intracellular gating), so they look more electropositive in this naive count than E1659A K1141 — which is a single Lys in an enclosed hydrophobic pocket, not a cluster.
The scan returns ARG439/LYS1104 for E1659A, NOT K1141, because K1141 has only hydrophobic neighbours (PHE1646, TRP1612, LEU1140 from druggability scan #1) and so doesn’t form a Lys/Arg cluster — the very thing that makes it a druggable pocket.
Why this is inconclusive without APBS
The two proxies compare different things and neither captures what Phase 5k measured:
- Phase 5k integrated the Coulomb potential at the formal-anion position over a 20-snapshot ensemble, capturing both charge density and geometric enclosure (the burial → high effective dielectric weight).
- Light-proxy charge counts ignore enclosure. Off-target K/R clusters in voltage sensors, etc., are surface-exposed and high-dielectric-screened; they would NOT show +5.99 kT/e in APBS, even with high charge density.
To match Phase 5k properly:
- fpocket (or my burial-grid estimator) on each off-target → enclosed pockets.
- APBS on each enclosed pocket → kT/e.
- Compare each to E1659A K1141 +5.99 kT/e.
This is the heavy planned step. The lite proxy cannot substitute.
Salvage
What we can claim from light data:
- Phase 8g Vina docking already ranked the lead’s affinity for these 4 off-targets vs E1659A. Vina ΔG was indistinguishable, which Phase 8g flagged as a Vina limitation.
- Phase 5k’s mech-4 mechanism is Coulomb-driven: the lead binds E1659A by anionic attraction to a single buried Lys (K1141). For an off-target to phenocopy this, it needs a single buried Lys/Arg in a small enclosed hydrophobic pocket. None of the 4 off-target structures presents an obvious such site by inspection of their published binding sites.
- This is suggestive but not quantitative. Wet-lab off-target panel (Phase 8 SOP gate 3) remains required.
Connections
[supersedes-attempt]STRC h01 Phase 8g v5.2 Off-Target Panel[blocks-on]Phase 9 APBS ensemble for off-targets (heavy)[ref]STRC h01 Phase 5k-WT Matched Ensemble APBS 2026-04-25