STRC Research

OpenMM

2 min read

OpenMM — molecular dynamics engine

What it does

Open-source MD engine (Python/C++). AMBER/CHARMM/OPLS force fields; explicit-solvent or implicit; Langevin/Verlet/NPT integrators; supports custom forces. Used here for Phase 4l free-energy / fibril dynamics.

Status

  • Installed: 2026-04-26, OpenMM 8.5.1.dev-f7fa0c2
  • Path: ~/STRC/hypotheses/h09-hydrogel/scripts/venv-md/bin/python (dedicated venv, isolates from system Python)
  • Install command: python3 -m venv venv-md && venv-md/bin/pip install openmm (Homebrew Python PEP 668 requires venv).

Apple Silicon path (M5 Max)

PlatformAvailableNotes
Reference✅speed=1.0; debugging only
CPU✅speed=10.0; Neon-vectorized; M5 Max 16-core perf cluster ~good
OpenCL✅speed=50.0 nominal; runs through Apple Metal under the hood
Metal (native)❌not yet in stable OpenMM; experimental openmm-mlx / openmm-ml exist
CUDA❌no NVIDIA on Apple Silicon

Practical guidance: for systems > ~10k atoms, OpenCL (Metal-backed) beats CPU. For tiny test systems, CPU wins due to GPU launch overhead. Phase 4l target system (~36k atoms RADA16-WH2 trimer + TIP3P water) should get ~30–100 ns/day on OpenCL.

h09 use case (Phase 4l, planned)

Refine C_eff parameter for the closed-form avidity model in STRC h09 Phase 4j Avidity Kd Model 2026-04-26 from Mammen-bracketed range (50–500 mM) down to a scaffold-specific number for THIS RADA16-WH2-tail91 construct.

Protocol sketch:

  1. Build 3 RADA16-WH2-tail91 monomers initially-aligned to mock fibril geometry (β-sheet H-bond pattern).
  2. Solvate (TIP3P, ~3 nm padding) + neutralize (NaCl 150 mM).
  3. Equilibrate NVT 100 ps → NPT 1 ns at 298 K, 1 atm.
  4. Production NPT 100 ns @ 2 fs.
  5. Analysis: WH2 motif center-of-mass time series → estimate WH2 angular distribution and inter-WH2 spacing on fibril surface → compute effective concentration C_eff via end-to-end PDF.

If C_eff ≥ 100 mM (Mammen rigid-scaffold mid-range) → Phase 4j conservative scenario closes → h09 promotes to full S.

Risks

  1. Force field for designed peptide: AMBER ff14SB or ff19SB for protein; standard. RADA16 not specifically parameterized but glutamate/lysine/arginine/alanine standard residues fine.
  2. β-sheet self-assembly stability over 100 ns: RADA16 self-assembly is templated by initial alignment. May drift if not constrained. Use weak position restraints for backbone H-bond donors/acceptors during equilibration; release for production.
  3. Sampling time: 100 ns may be insufficient for full fibril relaxation. Mitigation: replica exchange (REMD) or Hamiltonian-REMD if 100 ns deterministic is inconclusive.
  4. GPU compute time: 100 ns × 36k atoms on M5 Max OpenCL ~12–48h. Budget accordingly; not single-turn.

Connections

Graph View