announcementApr 7, 2026· SMA Research Platform
genmol_119_bbb_5 Stage 5 FAIL — Ligand Dissociates from LIMK2 in 20ns MD
#genmol_119_bbb_5#LIMK2#MD#Stage5#FAIL#negative_result
Result: FAIL
Holo molecular dynamics simulation (20ns, GAFF2, OpenMM on RTX 3090) shows genmol_119_bbb_5 completely dissociating from the LIMK2 ATP binding pocket.
Key Metrics
| Metric | Value | Gate | Verdict |
|---|---|---|---|
| Ligand RMSD | 106.7 Å | < 3 Å | FAIL |
| Ligand-protein distance | 33 → 79.5 Å | Stable | DISSOCIATED |
| Energy convergence | Stdev 1,287 kJ/mol | Stable | OK |
| Temperature | 300.3 K ± 0.9 K | 300 K | OK |
Root Cause
- Ligand placement bug: Fleet manager placed ligand at protein CoM (17.1, -3.5, 18.9), NOT at DiffDock-predicted ATP site (-13.2, 6.4, 28.0) — 33Å apart.
- DiffDock confidence not predictive of MD stability: +0.58 measures static pose quality, not dynamic binding.
- Stage 3b skipped: NeuralPLexer3 induced-fit refinement was not applied.
Control: Fasudil ROCK2 → PASS
Same MD setup, Fasudil stays in ROCK2 pocket (closest CA = 4.2Å). Methodology validated.
Raw data: github.com/Bryzant-Labs/sma-research/results/md_simulations/