STRC Research

STRC Pharmacochaperone Phase 4f Interface Rescue SMOKE

5 min read

STRC Pharmacochaperone Phase 4f Interface Rescue SMOKE

Phase 4f smoke test (1 lead only, indole-3-acetic-acid) completed 2026-04-22 06:04 UTC. Verdict is technically PASS but numerically meaningless. The static-snapshot interface ΔΔG method is not viable for this system — the absolute energies are too large relative to the real signal.

Input

  • Target: Ultra-Mini × TMEM145 full complex (job-ultramini-x-tmem145-full.cif, 18,771 atoms apo)
  • Lead: indole-3-acetic-acid (Phase 4b Vina ΔG = −6.05 kcal/mol, ranked #1 by raw affinity)
  • Method: OpenMM single-point + GBn2 implicit solvent
  • Iterations: APO 2×25, BOUND 8×25
  • Platform: CPU 18 threads (Apple M5 Max)
  • Wall clock: ~117 minutes (APO 31 min + BOUND complex 67 min + BOUND chain A 18.5 min)

Raw numbers

SystemAtomsE (kcal/mol)
APO complex18,771−9,773.6
APO chain A (solo)10,901−19,229.0
APO chain B (solo)7,870−4,413.7
BOUND complex + ligand18,792−33,184.8
BOUND chain A + ligand10,922−22,200.7

Derived:

APO  ΔG_interface = -9773 - (-19229) - (-4414)  = +13,869 kcal/mol
BOUND ΔG_interface = -33184 - (-22200) - (-4414) = -6,570 kcal/mol
ΔΔG_rescue        = APO - BOUND = +20,439 kcal/mol
recovery_fraction = 20439 / 8.4 = 2,433× (= 243,327% of 8.4 gap)
gate (>=30% recovery): **PASS**

Why the PASS is not real

Real pharmacochaperone binding energies are single-digit kcal/mol. A ΔΔG_rescue of 20,439 kcal/mol means the computation went off the rails — not that indole-3-AA is a miracle chaperone.

Diagnosis:

  1. Single-point GBn2 on 18,800-atom systems does not give clean absolute energies. The raw E values are ±10,000–20,000 kcal/mol each, and the interface ΔG is a difference of three such numbers. Any small geometric difference between APO and BOUND minima amplifies out to thousands of kcal/mol in the final ΔΔG.
  2. Adding a ligand (+21 atoms) and re-minimising pushes the system to a different conformational basin than the apo minimum. We’re not comparing like-with-like — we’re comparing two different local minima that happen to have a ligand in one of them.
  3. Short minimisation (200 iter on bound) does not converge to a physically meaningful structure. GBn2 in particular is known to need either long minimisation + restraints or MD-relaxed starting geometries.
  4. The reference gap of 8.4 kcal/mol (from STRC Electrostatic Analysis E1659A) is a real physical quantity. The computed ΔΔG of 20,439 kcal/mol is ~2,400× larger than any physically meaningful result. The ratio itself is the signature of numerical instability.

The script’s own header acknowledges this limit:

“For a clear pass or clear fail (>2 kcal/mol margin), it is decisive. For ambiguous results (within ±2 kcal/mol of the gate), Phase 5 short MD + dynamic MM-GBSA is the next step.”

Here the result is ~2,400× outside the “decisive” band. The method is giving garbage, not ambiguity. The “PASS” is a gate-boundary artifact, not a physics result.

What we actually learned

  • The Phase 4f static-snapshot pipeline does not work as designed for this system size. It was conceived for small-pocket ligand rescue, and Ultra-Mini × TMEM145 is a 19,000-atom two-chain complex with an interface ~hundreds of atoms.
  • Aspect (b) “loop-capping fold-stabilizer” mechanism from STRC Pharmacochaperone Phase 4c WT Decoy remains untested — Phase 4f as-built cannot answer it.
  • No meaningful signal for or against de-novo pharmacochaperone for STRC. Status unchanged from before Phase 4f: mechanism axis flagged, aspect (a) failed, aspect (b) untested, repurposing branch killed.

What would actually work

To get a real Phase 4f answer, need either:

  1. Phase 5 short MD + ensemble MM-GBSA: 100–500 ps equilibration + production MD with restraints, extract snapshots, compute MM-GBSA per snapshot, average. gmx (GROMACS) already installed. Wall clock: ~24–48 h on 18-core CPU, faster on GPU. Gets ±1 kcal/mol error bars on the ΔΔG_rescue number.
  2. MM-PBSA with explicit solvent (TIP3P): trade GB solvent model for explicit water; 50 ns MD + MM-PBSA. Longer wall clock, more rigorous.
  3. FEP+ with Schrödinger or Amber TI: commercial-grade free energy perturbation. Industry standard for pharmacochaperone lead optimisation but compute-heavy.

All three require Phase 5 MD infrastructure. Phase 5 was already marked as queued in STRC Pharmacochaperone Phase 4 Plan.

Don’t run Phase 4f production

The production run (3 leads) would take ~4 h wall clock and produce the same garbage for all three leads because the issue is methodological, not lead-specific. Canceled. The PHASE4F_INCLUDE_REPURPOSE=1 variant (which would have added 4PBA, IP-045, TMAO) is also moot — even if 4PBA docking mut-preference is real, the Phase 4f rescore would not detect it.

Ranking delta

STRC Pharmacochaperone Virtual Screen E1659A: no tier change, stays A (was demoted S→A earlier today on mechanism concerns; Phase 4f SMOKE provides no rescue). Evidence depth +1 on “method selection” axis: Phase 4f as-designed is not adequate; Phase 5 MD is the real next step. Next-step column updated.

No other hypothesis affected. STRC Repurposed FDA Chaperone Branch remains KILLED on mechanism grounds (independent of this result).

Connections

Graph View

Backlinks