from_openmm doesn't take System masses

[IAlibay]

Description

Reading in an OpenMM System with masses set should be reflected in the generated Interchange object (or maybe I'm looking at the wrong thing - sorry if that's the case!)

Reproduction

from openff.interchange import Interchange
from openff.toolkit import Molecule, ForceField
from openff.units import unit
import numpy as np
import os
os.environ['INTERCHANGE_EXPERIMENTAL'] = "1"

mol = Molecule.from_smiles("CC")
mol.generate_conformers()
sage = ForceField("openff-2.0.0.offxml")
cubic_box = unit.Quantity(30 * np.eye(3), unit.angstrom)

interchange = Interchange.from_smirnoff(topology=[mol], force_field=sage, box=cubic_box)
interchange.positions = mol.conformers[0]
omm_system = interchange.to_openmm(hydrogen_mass=4)
omm_top = interchange.to_openmm_topology()

interchange2 = Interchange.from_openmm(omm_system, omm_top)

mass1 = [at.mass for at in interchange.topology.atoms]
mass2 = [at.mass for at in interchange2.topology.atoms]
mass3 = [omm_system.getParticleMass(i) for i in range(omm_system.getNumParticles())]

print("mass1: ", mass1)
print("mass2: ", mass2)
print("mass3: ", mass3)

Output

mass1:  [<Quantity(12.01078, 'dalton')>, <Quantity(12.01078, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>]
mass2:  [<Quantity(12.01078, 'dalton')>, <Quantity(12.01078, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>, <Quantity(1.007947, 'dalton')>]
mass3:  [Quantity(value=3.0346209999999996, unit=dalton), Quantity(value=3.0346209999999996, unit=dalton), Quantity(value=4.0, unit=dalton), Quantity(value=4.0, unit=dalton), Quantity(value=4.0, unit=dalton), Quantity(value=4.0, unit=dalton), Quantity(value=4.0, unit=dalton), Quantity(value=4.0, unit=dalton)]

Suggested solution

I realised that both:

1. from_openmm is experimental
2. HMR within an interchange object probably going to break a ton of stuff

My suggestion would be to do a quick check when loading from openmm. If the masses don't agree then throw a user warning to tell folks that they just lost their non-standard masses.

Software versions

• Which operating system and version are you using? Ubuntu 20.04 LTS
• How did you install Interchange? conda-forge
• What is the output of running conda list? See JSON roundtrip doesn't work with from_openmm generated system #1003

[IAlibay]

Looks like it might be a duplicate of #888?

[mattwthompson]

Yes, this is (mostly) a duplicate. Unfortunately the best workaround I can suggest, on top of better warning/documenting this here, is just for the user to manage it themselves and remember about the HMR settings at export time.

If you're suggesting that the state of the Interchange should reflect the non-physical masses of the atoms, I agree that's a reasonable expectation but the toolkit doesn't allow masses to diverge from what their atomic number defines¹. I haven't thought through how a special representation could work, but it might be doable in the future if it's worth the diverging code paths.

Footnotes

1. This is a design I advocated for, forgetting about HMR at the time. Usually atomic masses are strictly defined by an atom's element, so it seemed like dead weight at the time. ↩

[IAlibay]

If you're suggesting that the state of the Interchange should reflect the non-physical masses of the atoms

Ah I should have probably clarified: as a user I generally think it's a good call for Interchange to be "physical things only". That sets clear expectations on what you can and can't do.