MT/HJ/JW Interchange check in

HJ prepared questions

• HJ – I notice different files coming out when I use OpenFF to parameterize an ethanol, and then I 1) write it to OpenMM, then use ParmEd to write that to GROMACS or 2) make an interchange and write to GROMACS

  • One issue is that the atom types are different between the two files

  • Another issue is that the GROMACS charge_ptype is listed as all 0s

    • MT – This is expected - The 0s are overwritten lower down in the file

• HJ – Planning to add support for alchemical topologies?

  • JW – We’d talked about it but it’s not planned for the next major release. I will keep you looped into any working groups that form on this. It wouldn’t be out any earlier than Q3 2022.

• HJ – I tried combining structures using the + operator in interchange, and it looks alright. Is this recommended for use?

  • MT – This should be considered experimental (and there’s a warning indicating as much), there’s a huge amount of complexity underneath all of it so we just need to keep testing it.

  • JW – Yes, if you are willing to help us, PLEASE do use the + operator and report bugs! This will be a huge help for us.

  • HJ – Gaetano had seen an issue using ParmEd where 1-4 scaling factors got mangled.

    • MT – What would be the expected behavior here?

    • HJ – I’d expect the energies to be additive (except for the inter-molecule energy, which would be new)

    • MT – Sorry, I think the 1-4 interaction energy is actually straightforward (in terms of what is “correct”). I was thinking of complexity/incompatibility for different mixing rules

• HJ – Will we be able to covalently attach proteins to small molecules?

  • JW – Plans are to basically append protein-specific parameters to Sage types and refit. This will be able to natively assign valence parameters to proteins, ligands, and modified proteins. But the big question is charge assignment - If we can get a neural net assigning AM1 charges then it should gracefully handle PTMs. But if not, then we’ll do library charges, with no support for PTMs. (or the user would need to somehow make up a librarycharge for their modified amino acid)

• HJ – Which Interchange version should I use for testing?

  • MT – 0.1.4, install from conda. This will change when the toolkit alpha package is available.

• HJ – Interchange.topology.to_openmm() is giving me blank OpenMM topologies

  • MT – Use interchange.topology.mdtop.to_openmm - We sneak an mdtraj topology along with the interchange Topology class. This will be changes in the alpha biopolymer package.

• HJ – Other requests for prioritization/timelines:

  • Biopolymers

    • JW – I’ll inform you when the alpha release of the infrastructure is ready. Anticipating Feb. We have a simple biopolymer FF (amber FF14SB) that can be used now. But Rosemary won’t be ready until late 2022 at the earliest.

  • Vsites

    • MT – Vsites are supported in interchange now, should be exported correctly in to_openmm, I’m not sure whether to_gromacs covers all of the types, and to_amber is not currently implemented.

    • MT – Are you thinking like extra sites on waters, or in general on small molecules?

    • HJ – I’d think small molecules

  • Custom functional forms (like anharmonic angles)?

    • MT + JW – You can implement that in OpenMM right now, would take about two hours to get something working. But for GROMACS export, we’re making a new plugin interface and we can add you to that working group. ETA on initial support is mid-2022.

• HJ – How should I give feedback?

  • JW + MT – The #developers channel on Slack. Also, let’s make this a regular check in.

  • (We will make this recurring, every week, but cancel if there’s nothing to discuss).