2021_12_01 Thompson/Wagner working session

Task planning

• Can move ahead with reference dataset, even if some details aren’t hammered out in the SMIRNOFF spec – Lots of peolpe just want to know if they’re getting torsions right, don’t care as much about precise agreement with long-range forces

  • MT will move ahead with reference energies dataset v1, listing assumptions that aren’t yet resolved by SMIRNOFF spec updates

• MT will look into making functionality to do capping+charging unrecognized AAs in a protein

  • LibraryChargeHandler._unrecognized_contiguous_subgraphs → list of tuple of atom indices?? Then we could have a way to turn those subgraphs into capped mols, assign charges, and turn them back into LibraryCharges??

  • May be good to get interface hints here from LWang’s polymeterizer

  • Assume that the ff is loaded as ForceField('openff-2.0.0.offxml','ff14sb_off_impropers_0.0.2.offxml'), and the user will call the _unrecognized... command chain between loading the FF and creating the system.

• Interchange importers?

  • MT – Blocked by lack of TypedMolecule, could use mdtraj top as a crutch for now.

  • MT – Concerned about scope creep. Some things are black and white - Loading a prmtop that uses ff03 should clearly be allowed, loading a prmtop wiht an unsupported vsite is clearly not our job (though folks are free to open PRs). But there’s a significant grey area around things like loading CMAPs from prmtop.

  • MT will work on AMBER loaders that can take “simple” systems, like

    • GAFF ligands

    • a protein that uses ff03

    • but not proteins with CMAPs until we know the full spec and whether different engines support them correctly

• Test usage of graph_tool instead of networkx for heavy comparisons in Toolkit

  • is/are_isomorphic → Relatively cheap, ok to be inefficient. Only expected to match entire molecules. SHOULDN’T try to match subgraphs. Less important to remove networkx here.

  • Molecule.from_pdb → Very expensive, could be sped up a lot. Needs to match subgraphs (individual residues in a larger chain). Important to find something faster than networkx here.

• Somehow improve SMARTS based parameter assignment for proteins. This would be really easy with multiprocessing, but that’s the last tool I want to reach for.

Units PR

• MT – When can we stop supporting simtk.units? It adds a good amount of complexity to openff-units.

  • JW – Let’s stop supporting it in the Feb release.

• JW will finish reviewing this on his own time – Logic looks good, will just act a s a second set of eyes for typos. If #1097 look good, we will close it and merge #1142, since it’s the same logic, just ported to the topology-refactor branch.

Topology performance/molecule grouping merge

• MT reviewed #1140, JW merged

Pushing molecule grouping updates into interchange

• MT will work on bringing similar performance improvements to those in #1140 into interchange

Topology-export-to-different-format needs

Resolving positions debate

• How many conformers can a molecule have?

  • One?

    • Direct correlation to simulation coordinates

  • Multiple?

    • Allows ELF10 work

  • Unresolved - Opened to debate on #developers channel

• Can a Topology have conformers?

  • No

• What is the relationship between interchange positions and molecule conformers?

  • Interchange positions will be totally separate