Caution

This code is still in development. Visit https://github.com/Hori-Lab/UNISIS for the latest version of our model.

OpenMM scripts to run UNISIS RNA simulations

Installation

OpenMM environment

OpenMM can be installed by mamba/conda.

(torchmd) $ mamba install openmm

It is also pre-installed on the Pharmacy HPC.

[bluto:] $ module load openmm

Install from source

git clone https://github.com/Hori-Lab/unisis-omm.git
cd unisis-omm
pip install -e .

How to run simulations with a neural-network potential (NNP)

Install OpenMM-Torch

To use NNP trained by TorchMD-net, we use OpenMM-Torch.

Again, this package can be easily installed by mamba.

(torchmd) $ mamba install openmm-torch

Simulation set up

To run simulations, you need

• model file (e.g. epoch=299-val_loss...)
• force field file (e.g. htv23_nnp_dihexp.ff)
• simulation input YAML file (e.g. simulate.yamll)
• PDB file for the initial structure (e.g. T2HP_unfolded.pdb)

Example set up are in examples/TMnet_T2HP/.

Note that, in the PDB file, residue names have to be RA, RU, RC, RG, and RD (terminal dummy).

For the input YAML file, you can use almost the same input file as for the TorchMD input. You just have to add the following line for the initial structure PDB (not DCD).

structure: ./T2HP_unfolded.pdb

Relevant lines for the OpenMM script are:

timestep: 50
langevin_gamma: 0.5
langevin_temperature: 300
temperature: 300
output: md
output_period: 1000
save_period: 1000
seed: 10001
steps: 5000000
structure: ./T2HP_unfolded.pdb
external:
  embeddings: [5, 2, 3, 4, 1, 4, 2, 1, 2, 2, 4, 3, 1, 4, 1, 3, 1, 4, 3, 2, 4, 3, 1, 4, 1, 2, 3, 1, 5]
  file: ./epoch=213-val_loss=0.0335-test_loss=0.0690.ckpt

All other lines will be ignored in OpenMM job. (See # ignored comments in simulate.yaml.

Running the simulations

Submit your job by the following command.

(For GPU)

unisis-omm  --tmyaml simulate.yaml \
            --ff htv23_nnp_dihexp.ff \
            --cuda \
            1> out 2> err

(Otherwise)

unisis-omm  --tmyaml simulate.yaml \
            --ff htv23_nnp_dihexp.ff \
            1> out 2> err

An example HTCondor script is condor.txt.

Restarting a simulation

You can restart a simulation using *.rst file by adding -r or --restart option in the input command.

unisis-omm --tmyaml simulate.yaml \
           --ff htv23_nnp_dihexp.ff \
           --restart ./run1/md.rst \
           --cuda \
           1> out 2> err

Important
If you restart the simulation in the same directory using the same input file, existing output files such as md.out and md.dcd from the previous run will be overwritten, i.e. you will lose the data! Therefore you must either change the prefix of the output filename or move previous results to another directory.

DCD and DCD_FORCE modes

Use DCD or DCD_FORCE mode to re-calculate energy and force from existing DCD file.

You just need to add the following two lines to the YAML input file.

(DCD mode)

job: DCD
dcd_file: ./run1/md.dcd

(DCD_FORCE mode)

job: DCD_FORCE
dcd_file: ./run1/md.dcd

Important
If you run DCD/DCD_FORCE mode in the same directory using the same input file, existing output file md.out will be overwritten, i.e. you will lose the data! Therefore you must either change the prefix of the output filename or move previous results to another directory.

In DCD mode, only energies are calculated and output PREFIX.out file, where PREFIX is specified as output in the YAML input file. In DCD_FORCE mode, energies and forces are calculated and forces will be writtein in separate files.

References