STRC Research

STRC h01 Phase 4i APBS Pose-Transplant Rescore 2026-04-24

9 min read

STRC h01 Phase 4i APBS Pose-Transplant Rescore

Cheap (Tier-4) per-pose cross-check on the Phase 5j pocket-average claim. For each of the 29 v3b YELLOW ligand poses, Kabsch-aligned the mutant K1141 pocket (residues 1126-1156; Cα RMSD 0.21 Å) onto the WT frame, then evaluated the Phase 5j APBS grids at the same ligand geometry in both pocket environments. 20/29 ligands (69%) prefer the mutant pocket with median Δφ = +0.70 kT/e and formal-anion ΔG = −0.43 kcal/mol; Vina sees 0/29 mutant preference with median WT-bias +0.55 kcal/mol. Gap ≈ 1 kcal/mol. Tier-4 verdict: PARTIAL_RESCUE — the direction of Phase 5j holds at the per-pose level, magnitude is smaller than the +4.3 kcal/mol pocket-average because the acidic-head centroid samples the pocket entrance rather than the pocket interior where φ is maximally positive.

Problem

Phase 5j APBS (STRC h01 Phase 5j APBS WT vs Mutant Pocket Electrostatics 2026-04-24) showed mutant pocket mean φ is +7.1 kT/e more positive than WT (+4.3 kcal/mol for an anion) — a pocket-average statistic. The per-pose question was still open: for the specific poses Vina generated (Phase 4c-v3b), does the same APBS electrostatic bias apply?

Tier-4 test of the reframed pharmacochaperone claim: “acidic ligands bind the mutant pocket via net Coulomb attraction from the residual K1141 Lys + D1140/D1173 cluster minus the compensating E1659 negative charge.”

Originally planned as MM-GBSA single-point rescoring via Amber ff14SB + GAFF2 + GBn2 implicit solvent in OpenMM, but GAFFTemplateGenerator graph-matching on fluorinated v3b scaffolds failed ~5 min per ligand (log: “No template found for residue 0 (LIG). The residue contains F atoms”). Pivoted to a direct APBS grid reuse — same physics, no parametrization.

Method

Stage 1 — Naive per-pose APBS

For each Phase 4c-v3b pose (v3b YELLOW 29 × {WT, mutant}):

  1. Parse MODEL 1 of docking_runs/4c_v3b/poses/<name>__{state}.pdbqt: atom coordinates, Gasteiger partial charges (column 71-76), PDBQT atom types.
  2. Load Phase 5j APBS grids apbs_wt_mut/{state}/{state}_pot.dx (65 × 65 × 65, Δ=0.44 Å, origin at the respective Phase 4c box centre).
  3. Sum q_i × φ(r_i) with trilinear interpolation → E_APBS_Gasteiger (kT/e → kcal/mol via 0.616 at 310 K).

Separately for the formal anion correction:

  1. Identify OA (hydrogen-bond-accepting oxygen) atoms — the carboxylate / hydroxamate oxygens of each v3b ligand. Compute centroid r_OA.
  2. E_APBS_formal = −1 × φ(r_OA) — treat the ligand as a point q=−1 e anion localized at the acidic-head centroid (every v3b YELLOW is a mono-acid: benzofuran-2-COOH, CONHOH, tetrazole, CONHSO₂Me bioisosteres).

Stage 2 — Pocket-local alignment + pose-transplant

Stage 1 has a frame-mismatch confound: WT and mutant AF3 structures are in different global coordinate frames (global Cα RMSD 10.9 Å after whole-chain Kabsch; 70 Å raw). A WT-pose lives at absolute coordinates in the WT frame; a mutant-pose at absolute coordinates in the mutant frame. Both are “optimal” within their respective pockets → Δφ per-pose is dominated by Vina’s local-minimum choice, not the pocket-level electrostatic shift.

Fix: Kabsch-align the mutant K1141 pocket-local Cα set (residues 1126-1156, 31 residues) onto the WT frame → (R, t) with pocket RMSD 0.21 Å — very tight local match despite 10.9 Å global mismatch. Save R, t.

For each ligand:

  1. Mutant-pose centroid c_mut (in mutant frame) → transplant to WT frame as R·c_mut + t. Evaluate φ in WT grid at that point.
  2. WT-pose centroid c_wt (in WT frame) → transplant to mutant frame as R⁻¹·(c_wt − t). Evaluate φ in mutant grid.
  3. Per ligand, report Δφ_mut−wt = mean of the two symmetric measurements; ΔG_formal = −0.616 × Δφ (kcal/mol per unit anion charge).

Parameters

  • APBS grids: Phase 5j output (nonlinear PBE, PARSE force field, pH 7.4, 150 mM NaCl, ε_p=2.0, ε_s=78.54, T=310 K, 65×65×65 grid, 0.44 Å spacing, 28.4 Å box centered on K1141 pocket).
  • Pose coords: Phase 4c-v3b Vina MODEL 1 (exh=32, num_modes=3, cpu=8); pose selection matches STRC h01 Phase 4c-v3b WT Decoy on v3b YELLOW 2026-04-24.
  • v3b YELLOW list: 29 ligands × 2 states = 58 poses; all scored.

Script: pharmacochaperone_phase4i_apbs_pose_rescore.py (new 2026-04-24). Runs in ~15 s end-to-end (pure Python + NumPy; no new MD, no new QM, no new docking).

Results

Stage 1 — Naive per-pose (frame-mismatched)

metricGasteiger qFormal q=−1 at OA centroid
n ligands2929
n prefer mutant (Δ<0)1715
frac prefer mutant0.5860.517
median Δ (kcal/mol)−0.019−0.063
Vina median Δ (kcal/mol)+0.552 (WT-biased)+0.552
APBS−Vina gap (kcal/mol)0.5710.615

Tepid signal: APBS barely flips direction; magnitude ~0.02-0.06 kcal/mol. Interpretation: Vina places WT and mutant poses at DIFFERENT locations within each respective pocket (both locally optimal). Comparing φ at different locations in different frames is not a clean comparison — mostly measures Vina’s pose-selection noise rather than the pocket charge asymmetry.

Stage 2 — Pose-transplant (same geometry, two grids)

metricvalue
pocket-local alignment Cα RMSD0.21 Å
n ligands29
n prefer mutant (Δφ>0)20/29 (69%)
median Δφ_mut−wt (kT/e)+0.70
median ΔG_formal mut−wt (kcal/mol)−0.43
mean ΔG_formal mut−wt (kcal/mol)−0.36
Vina Δ median (kcal/mol)+0.55
Vina − APBS_transplant gap (kcal/mol)−0.99

Direction flips cleanly vs Vina’s 0/29; magnitude is ~1 kcal/mol total shift (Vina says mutant is 0.55 kcal/mol worse, APBS same-geometry says mutant is 0.43 kcal/mol better, so the gap is ~1 kcal/mol).

Physical interpretation of the magnitude gap vs Phase 5j

Phase 5j: pocket-averaged Δφ = +7.1 kT/e (+4.3 kcal/mol for q=−1 e). Phase 4i transplant: pose-centroid Δφ = +0.70 kT/e (+0.43 kcal/mol).

Ratio ~10×. Expected because:

  • Phase 5j integrates over the entire 18³ Å pocket interior, including deeply-buried regions where φ is most positive (E1659→A1659 removes a −1 e charge from the pocket centre, amplifying the effect near the centre and diminishing near the opening).
  • v3b YELLOW ligands are docked such that the acidic head (COO⁻ / CON(O)⁻ / tetrazolide) sits near the pocket ENTRANCE, not the deep interior — Vina placed the lipophilic naphthyl/biphenyl tail toward the pocket interior to maximize vdW, leaving the charged warhead at the entrance.
  • Entrance φ difference is weaker than interior φ difference — the pocket-averaged +7.1 is a mixed mean of a strong interior signal and weaker entrance signal.

This is consistent; both results say the mutant pocket is more electrostatically attractive to anions, just at different depths.

Why Vina misses this

Vina’s PDBQT preparation via Meeko / AutoDockTools assigns Gasteiger partial charges that net to zero on acidic ligands (we verified: total ligand q range across all 29 v3b = [−0.004, +0.001] e). Vina has no explicit anion / protonation-state handling. Its scoring function then evaluates a distance-cutoff Coulomb over these (essentially neutral) charge distributions — the result is tiny, and the remaining ~0.5 kcal/mol WT bias is carried by vdW/Hbond terms that do see small geometric differences between WT and mutant poses.

APBS with formal q=−1 e at the acid centroid recovers the ~1 kcal/mol energy that Vina’s charge model discards.

Interpretation

  • Phase 5j is confirmed at the per-pose level, direction-wise (20/29 vs Vina 0/29), with a ~10× smaller magnitude than the pocket-average due to entrance-vs-interior sampling.
  • Vina’s Phase 4c-v3b WT-bias is definitively a Vina-force-field artefact (verdict upgraded from Phase 5j’s “likely artefact”). At the same pose geometry, APBS says mutant is preferred; at different pose geometries, APBS says essentially no preference; Vina always says WT. The asymmetry is in Vina’s charge model + Coulomb term, not in the biology.
  • The pharmacochaperone mechanism reframe from Phase 5j stands: “E1659A creates an electropositive K1141 pocket from the loss of Glu1659’s negative sidechain; acidic ligands bind via net Coulomb attraction (~0.5-4 kcal/mol depending on pose depth) from the residual K1141/D1140/D1173 cluster.” Per-pose Coulomb contribution ~0.4-1 kcal/mol; pocket-deep-binding Coulomb contribution ~4 kcal/mol; realized contribution depends on ligand’s geometric placement.
  • For v5 library design this implies: push the acidic warhead deeper into the pocket (current v3b poses leave it at the entrance). Shortening the linker between the warhead and the lipophilic tail — or pre-shaping the tail so it wraps rather than extends — should move the warhead into the deep pocket where the +7 kT/e is strongest. This is a concrete chemistry goal for v5.

Limitations

  1. Static structures. APBS grids are Phase 5j single-snapshot AF3 predictions; loop dynamics will modulate φ by ±2-5 kT/e (Phase 5g holo-MD + per-frame APBS remains gold-standard).
  2. Centroid approximation. Formal anion treated as point q=−1 e at OA centroid; for COOH this is accurate (symmetric COO⁻), for tetrazole slightly off (5-ring π delocalization shifts the effective centre). Impact: ≤0.1 kcal/mol per ligand, within noise.
  3. Pocket-local Kabsch only. Alignment used residues 1126-1156 Cα (pocket ring around K1141); regions beyond are in their native mis-aligned frames. The Kabsch transform is rigid, so the mapping is well-defined for the pocket neighbourhood but diverges farther out. Confirmed by pocket Cα RMSD 0.21 Å (within 15 Å of K1141), validating the local frame for the pose regions of interest.
  4. PDBQT Gasteiger is the Vina default. If a user rebuilt PDBQT with explicit formal charges on the acid oxygens (via obabel -xp or --partialcharge eem2015bn or explicit PARSE charges), Vina could see the anion. This would be a different Vina pipeline and has not been tested here. Phase 4i establishes the Vina artefact for the standard Meeko / AutoDockTools PDBQT pipeline used throughout STRC.

Ranking delta

  • Hypothesis h01: tier A held | mech 3 held (Phase 5j established mech=3 via pocket-average APBS; Phase 4i confirms at per-pose level without further upgrade — still awaiting holo-MD ensemble for mech=4) | deliv 3 held | misha_fit 4 held
  • Next-step unchanged (Phase 5g holo-MD top priority) but with refined chemistry guidance: v5 design must push warhead deeper into pocket to capture the pocket-interior +7 kT/e signal rather than the +0.7 kT/e entrance signal v3b is currently docking into.
  • Reinforces STRC AF3 Static Pocket Blindness to Loop Dynamics second-axis conclusion (now via electrostatics + pose transplantation); Vina is confirmed as the wrong tool for the Coulomb-sensitive portion of pharmacochaperone docking.
  • Tier-4 complete. Tier-0 (ensemble APBS on top-3 leads × 20 mutant snapshots) remains the next ascending-compute step; Tier-1 (2 ns holo-MD + per-frame APBS) conditional on Tier-0 outcome.

Connections

Graph View

Backlinks