MD Binding-Energy Leads
COMPUTATIONAL HYPOTHESISSingle-trajectory MM-GBSA — flagged PREMATURE
The ΔG values below are single-trajectory MM-GBSA (gbn2, 100 frames, no entropy). They rank-order candidates within a single target only and are not absolute binding affinity. A multi-LLM review consortium (Claude + GPT + Gemini) flagged this candidate set PREMATURE pending a decoy + positive-control benchmark that has not yet completed. Every row is a computational hypothesis, not a wet-lab-validated lead, and nothing here is an efficacy claim. Two candidates are flagged and de-ranked: a MAPK12 value driven by molecular size (ligand-efficiency artifact) and a MICU1 candidate carrying an aromatic-nitro toxicophore.
Each row is a 100 ns OpenMM molecular-dynamics run on an 8×H100 node against a canonical SMA priority target, with an MM-GBSA end-state binding-energy estimate, the backbone RMSD (trajectory stability), the AutoDock Vina docking score, and the multi-method ligand consensus where available. Results land directly to canonical storage (Spark1).
▶How does MM-GBSA binding-energy estimation work?
ΔG (MM-GBSA). Molecular Mechanics / Generalized-Born Surface-Area end-state free-energy estimate (gbn2 model, 100 trajectory frames, no entropy term, CPU). More negative = stronger predicted binding within the same target. Because the estimate has no entropy correction and a single trajectory, it is a ranking signal, not an affinity; absolute values are not comparable across targets and can track ligand size.
MD-RMSD (nm).Backbone root-mean-square deviation over the trajectory — < 0.6 nm = stable (green), 0.6–1.0 nm = moderate (amber), > 1.0 nm = unstable. Note: backbone stability is not the same as pose stability.
Vina. AutoDock Vina docking score (kcal/mol) for the same compound/target (≤ −8 indicates a strong docked pose). Consensus. Multi-method ligand agreement (Vina + Boltz-2 + Chai-1) where computed.
Why PREMATURE. The consortium required binder/decoy discrimination (decoys must score measurably worse than candidates) before any wet-lab framing. That benchmark is not yet complete, so no row here is actionable as a lead.