[WIP] Overhaul parameterization scheme used for small molecules

[jchodera]

This PR (originally #517) changes the way that we parameterize small molecules with GAFF.

OEGAFFTemplateGenerator

Instead of using the flawed gaffTemplateGenerator, which was buggy in perceiving chemistry from the OpenMM Topology (which lacks chemical information)---and didn't even run in many cases on osx---I have implemented a new OEGAFFTemplateGenerator that accepts OEMol versions of molecules it should match within the Topology object in order to determine what chemical species they are.

The OEGAFFTemplateGenerator also supports the ability to specify a JSON cache file, allowing parameters for small molecules to be computed only once and then cached thereafter. This should also greatly speed up testing.

Examples:

Create a template generator for GAFF for a single OEMol and register it with ForceField:

    >>> from openmoltools.forcefield_generators import OEGAFFTemplateGenerator
    >>> template_generator = OEGAFFTemplateGenerator(oemols=oemol)
    >>> from simtk.openmm.app import ForceField
    >>> forcefield = ForceField('gaff.xml', 'amber14-all.xml', 'tip3p.xml')
    >>> forcefield.registerTemplateGenerator(template_generator.generator)

Create a template generator for GAFF2 for multiple OEMols:

    >>> template_generator = OEGAFFTemplateGenerator(oemols=[oemol1, oemol2], gaff_version='gaff2')

You can also some OEMols later on after the generator has been registered:

    >>> forcefield.add_oemols(oemol)
    >>> forcefield.add_oemols([oemol1, oemol2])

You can optionally create or use a tiny database cache of pre-parameterized molecules:

    >>> template_generator = OEGAFFTemplateGenerator(cache='gaff-molecules.json')

Newly parameterized molecules will be written to the cache, saving time next time!

SystemGenerator

The SystemGenerator has been moved to perses.forcefield. To use the OEGAFFTemplateGenerator, we simple specify some oemols and optionally a cache path to the JSON file that stores small molecule parameters. This will automatically configure an OEGAFFTemplateGenerator.

Here's an example:

    # Set up SystemGenerator for vacuum or explicit solvent
    gaff_filename = get_data_filename('gaff.xml')
    forcefield_ffxml_filenames = [gaff_filename, 'amber99sbildn.xml', 'tip3p.xml']
    nonbonded_method = app.PME if solvent else app.NoCutoff
    barostat = openmm.MonteCarloBarostat(1.0*unit.atmosphere, temperature, 50) if (use_barostat and solvent) else None
    cache = get_data_filename('OEGAFFTemplateGenerator-cache.json')
    default_forcefield_kwargs = {'removeCMMotion': False, 'nonbondedMethod': app.NoCutoff, 'constraints' : app.HBonds}
    forcefield_kwargs = default_forcefield_kwargs.update(forcefield_kwargs) if (forcefield_kwargs is not None) else default_forcefield_kwargs
    system_generator_kwargs = system_generator_kwargs if (system_generator_kwargs is not None) else dict()
    system_generator = SystemGenerator(forcefield_ffxml_filenames,
        forcefield_kwargs={'removeCMMotion': False, 'nonbondedMethod': nonbonded_method},
        barostat=barostat,
        oemols=[old_oemol, new_oemol],
        cache=cache, **system_generator_kwargs)

Status

I'm still working on remediating code and examples that had previously used the gaffTemplateGenerator.

• Add tests for SystemGenerator
• Update perses/tests/testsystems.py
• Update perses/tests/test_RJMC_vacuum.py
• Update perses/tests/test_hybrid_builder.py
• Update perses/tests/test_elimination.py
• Fix issues in perses/dispersed/feptasks.py related to managing nonequilibrium integrators attached to contexts and those not attached to contexts

[jchodera]

I should be able to finish this up quickly now!

[jchodera]

@hannahbrucemacdonald @dominicrufa : Would be great to get your reviews / comments on this, since I think I've resurrected all the tests I can easily resurrect at this point.

Be sure to check out the change in information flow where OEMols are now passed around instead of constructing them from OpenMM Topology objects (which was highly unreliable).

[jchodera]

Supplanted by #616 and #620