RAMD-OpenMM

OpenMM plugin implementing τRAMD (Random Acceleration Molecular Dynamics) for relative ligand residence-time ranking.

What it is

Applies a stochastic outward force on the ligand centre of mass during MD; force direction reseeds when displacement falls below a threshold. Measures time-to-unbind across replicas. Returns relative residence-time ranking (τ_RAMD) that has been shown to rank-correlate with experimental k_off across diverse target classes (Kokh 2018).

Not absolute k_off. For absolute k_off use metadynamics + TST or Markov state model from long unbiased MD.

Why H01 needs it

Every H01 affinity number to date (Vina ΔG, MM-PBSA, APBS, Phase 8h-lite pose stability, Phase 7I Boltz-2 ipTM) is equilibrium-flavoured. For a pharmacochaperone the success metric is residence time, not K_d (see Pharmacochaperone Residence Time Criterion). τRAMD on the v5.2 shortlist closes that gap as Phase 5m.

Inputs

• Receptor + ligand pose (recommend Phase 5l K1141 fragment, not full-length 1775-aa system — 4× faster MD per STRC K1141 Fragment Construct Strategy)
• OpenMM-compatible force field (Amber sb14 + GAFF2 / OpenFF for ligand)
• Initial pocket COM, unbinding threshold (typ. 8 Å beyond initial COM)

Tunable parameters

• Force magnitude — typ. 14 kcal/(mol·Å); too low → no unbinding within 5 ns, too high → poses rip apart non-physically.
• Reseed interval — typ. 100 ps if displacement < 0.025 Å.
• Replicas — 10-20 per ligand for SEM.
• Max sim time — 5-10 ns per replica.

Cost

10 ligands × 10 replicas × 5 ns = 500 ns total wall. At Phase 5l fragment scale (~50 ns/day Metal-OpenMM on M-series Mac) → ~10 days serial, or split as 4× parallel replicas → ~2.5 days. Manageable as overnight × 3.

Install

pip install openmm openff-toolkit openmmforcefields
pip install git+https://github.com/seekrcentral/openmm_ramd.git

The actual canonical OpenMM port of τRAMD is at github.com/seekrcentral/openmm_ramd (the older HITS-MCM/ramd_openmm URL cited in the Kokh 2018 supplement is dead — repository deleted). Verified install 2026-04-26 in strc-mmgbsa conda env (openmm 8.5.1, openff-toolkit 0.18.0).

API: from openmm_ramd.openmm_ramd import RAMDSimulation. Constructor signature:

RAMDSimulation(
    topology, system, integrator,
    ramd_force_magnitude,        # openmm Quantity in kcal/(mol·Å)
    ligand_atom_indices,         # iterable[int]
    receptor_atom_indices=None,
    ramdSteps=50,                # MD steps between force re-direction checks
    rMinRamd=0.025,              # nm — re-direction threshold (≈ 0.25 Å)
    maxDist=50,                  # nm — unbinding criterion (50 Å beyond start)
    forceOutFreq=50,
    platform=None,
    logFileName=None,
)

Use sim.run_RAMD_sim(maxSteps) for production; sim.RAMD_step reports current step.

Reference

Kokh DB et al. Estimation of drug-target residence times by τ-random acceleration molecular dynamics simulations. J Chem Theory Comput. 2018;14(7):3859-3869. doi:10.1021/acs.jctc.8b00230

Connections

• [part-of] STRC index
• [implements] Pharmacochaperone Residence Time Criterion
• [depends-on] STRC K1141 Fragment Construct Strategy