Openrid: Reinforced Dynamics Built from OpenMM & PyTorch

WARNING: Packages are still under development and not well tested. Please use with caution.

Features

• Parallel tempering sampling for different walkers
• At least 20% improvement compared to Gmx+TF
• Easy installation, one command

Quick Start

There is an example in example folder, run the script below to see if it works. You can modify the ala2.yaml file to change the parameters.

cd example
python run.py

Setup

mamba create -n rid_openmm -c conda-forge openmm-torch=1.1 pytorch=2=*cuda* parmed MDAnalysis mpi4py mpich mpi clusterutils openmmtools

If GLIBC doesn't work or you are working on SLURM cluster, try:

CONDA_OVERRIDE_CUDA=11.8 CONDA_OVERRIDE_GLIBC=2.35 mamba create -n rid_openmm -c conda-forge openmm-torch=1.1 pytorch=2=*cuda* parmed MDAnalysis mpi4py mpich mpi clusterutils openmmtools

If mamba doesn't work, change to conda install/create.

installation

python setup.py install

Speed Benchmarks

Tested on Chignolin system (one V100 GPU card)