I don't think you want to include charge_from_molecules here, do you? You want ff to assign AM1-BCC charges, which is specified as part of the openff force field you are using as a base force field (here, openff-1.2.0).

after calling assign_partial_charges() the charges would be stored in that molecule?

Ah, I see! Perhaps we need am1-bcc, which calls assign_partial_charges(), and a separate smirnoff options, which just leaves this up to the ff. object to handle charges as specified in the force field?

Doesn't assign_partial_charges require a partial_charge_method argument to be specified?
https://open-forcefield-toolkit.readthedocs.io/en/0.10.1/api/generated/openff.toolkit.topology.Molecule.html#openff.toolkit.topology.Molecule.assign_partial_charges

Here, you do want charge_from_molecules, which will override the charging scheme and take the charges from g.mol.

yes but this is assuming that the molecule has charges so that we don't need to redo?

Yes, if this is specified, we are assuming the molecule already has charges.

Can you document charge_method in the docstring, and maybe add something to the README.md?

I think we support

    charge_method : str, optional, default='nn'
        Method to use for assigning partial charges:
        'nn' : Assign partial charges from the espaloma graph net model
        'am1-bcc' : Allow the OpenFF toolkit to assign AM1-BCC charges using default backend
        'gasteiger' : Assign Gasteiger partial charges (not recommended)
        'keep' : Use partial charges provided in the original `Molecule` object

Perhaps we could rename

• keep -> from-molecule
• nn -> espaloma to use "espaloma" charges?

What does the else: behavior bellow do? Shouldn't we raise an exception for the else: case because the charge method is not recognized?

for else it would use the default behavior of ff.create_openmm_system which is to use am1-bcc as well?

I think we want to avoid else:, since we don't want to have typos in charge_method silently give weird behavior---typos in charge_method should raise a ValueError.