2022-06-10 OpenMM/OpenFF Check-in meeting notes

Update on F@H interface project

@Jeffrey Wagner

• Aiming for completion in Sep (Chodera/AVID goals, noneq cycling) and Oct (OpenFF goals, noneq switching)

• Discussing data formats. Ideal would be single PDB for each protein that can be read and parameterized by OpenMM and GROMACS (with AMBER as a silver medal)

Building topologies with OpenFF

Peter Eastman

• PE – Getting ready to put out openmm 8 beta. One big focus is ML/NN potentials. For example, could use AMBER for protein and NN for interal energy of ligand. So one issue is that there aren’t many good NN potentials now, and the lack of datasets to make it. But ultimately we want a simple API for people to use this. Right now we plan to support ANI1 and 2, but we want to extend it in the future.

  • Right now the input is an OpenMM topology. ANI only needs (element and coordinates). But we anticipate that other methods we’ll want to support will need (element, formal charge, bond orders). So we’re looking into how to extend the API as to get this information. One promising approach is the OpenFF toolkit molecule. But I’m interested in more details.

  • JW –

    • Molecule I/O

      • PDB for small molecule will be hard

      • PDB for proteins (and eventually other polymers) will be supported.

        • Active development (0.11.0 pre-alpha, colab notebook exists somewhere for testing)

        • Current focus is “vanilla” proteins, uses CIF structure database from RSCB

          • Does not trust atom names of input PDBs, just looks at elements and connectivity before doing substructure matching

        • Does not support post-translational modifications

        • Later releases MAY support other biopolymers (nucleic acids, less likely to be sugars later)

    • Information content

      • OFFMol proteins will have

        • atoms: element, formal charge, stereochemistry (conditional), aromaticity

        • bonds: bond order, stereochemistry (conditional), aromaticity

        • Optional metadata field (offmol.properties)

        • High-quality roundtrips with RDKit, atom ordering is guaranteed to be preserved.

• PE – Keep in mind that OpenFF’s PDB loader WILL be dependent on correct atom naming, by virtue of using OpenMM’s PDBFile to load them. PDBFile can only fill in the CONECT information if the residues have the correct atom names.

Confirm thoughts on point polarizability

@Jeffrey Wagner

• A FF research project wants to do simulations with point polarizability in an external electric field. It looks like a CustomNonbondedForce can be implemented to have an external field, and AMOEBAForces can do point polarizability, but there’s no straightforward way to combine them.

  • looks like another implementation of polarizability, but which also doesn’t allow external E fields.

  •  

• PE – This is basically impossible/very very hard. Or, it may be possible to hack this into AmoebaMultipoleForce. It would take deep modifications to the code and they should open an issue on GitHub to discuss.