molssi / mmic_openff

Tactic MMIC translator for OpenFF/MMSchema

OpenFF translator for MMSchema

This is part of the MolSSI Molecular Mechanics Interoperable Components (MMIC) project. This package provides translators between MMSchema and OpenFF toolkit.

mmic_openff provides 2 classes of translators for: molecules and forcefields.

Models

from mmic_openff.models import OpenFFMol

# Convert MMSchema to OpenFF molecule
off_mol = OpenFFMol.from_schema(mm_mol) -> 

# Convert OpenFF to MMSchema molecule
mm_mol = OpenFFMol.to_schema(off_mol) -> mmelemental.models.Molecule

One could do similar conversions for the ForceField model as well.

Components

The from_schema and to_schema methods in the OpenFFMol model use translation components provided by mmic_openff and MMElemental to convert between MMSchema and OpenFF representations.

from mmic_openff.components import OpenFFToMolComponent, MolToOpenFFComponent
from mmic_openff.models.import OpenFFMol
from mmelemental.models import Molecule

MMSchema to OpenFF molecule

# Create MMSchema molecule
mm_mol = Molecule.from_file(path_to_file)

# Create translation input
inp = {
    "schema_object": mm_mol,
    "schema_version": 1,
    "schema_name": "mmschema",
}

# Run translator compute
outp = MolToOpenFFComponent.compute(inp)

# Extract OpenFF molecule object
mol = outp.data_object.data

OpenFF to MMSchema molecule

from simtk.openmm import app

# Create OpenFF input

...

# Create translation input
inp = {
    "data_object": ommol,
    "schema_version": 1,
    "schema_name": "mmschema",
}

# Running translator compute
outp = Translator.compute(inp)

# Extract MMSchema molecule
mm_mol = outp.schema_object

One could do similar conversions with the force field component as well.

Copyright

Copyright (c) 2021, MolSSI

Acknowledgements

Project based on the Computational Molecular Science Python Cookiecutter version 1.5.

mmic_openff's Issues

Dynamic/static units

The mol and ff converters assume that any unit used for a particular variable in the OpenFF schema remains constant throughout. This assumption is made for the sake of code clarity and better performance. Also MMSchema does not support dynamic units for the same field. Should it?

Example:

>>> smirnoff_data["ProperTorsions"]["Proper"][0]["k1"]
    '0.1703506220295 * mole**-1 * kilocalorie'
>>> smirnoff_data["ProperTorsions"]["Proper"][-1]["k1"]
    '0.0 * mole**-1 * kilocalorie'

The converters extract the units for k1 in this case from smirnoff_data["ProperTorsions"]["Proper"][0] and assume it is the same for the [-1] entry.

Multiple conformers

Currently MMSchema does not handle multi-conformers (whereas OpenFF toolkit does). What is the best way to handle this disrepancy?

1 - Add another dimension to Molecule.geometry ? This would make Molecule overlap with Trajectory to some degree but would make Molecule a relevant output to equilibrium MM computations as well. It seems like a neat solution.
2 - Have mmic_translator optionally return multiple Molecule objects? This would make the process of translation between MMSchema and other codes/specs a bit awkward and less rigorous.
3 - Populate the Molecule.extras field with any extra conformers? This is a temporary solution but may last a while as long as every MM code does things in its own way.