09-20-2022 Thompson/Wang Benchmarking Check-in

Owned by Matt Thompson, created
Last updated: Sept 20, 2022
1 min read

Participants

  • @Matt Thompson

  • @Lily Wang

  • @Jeffrey Wagner

Goals

  •  

Discussion topics

Item

Presenter

Notes

Item

Presenter

Notes

Go through updated code

@Matt Thompson

    • Minimal topology/molecule, force field, and system layers exist

    • Can great side-by-side OpenMM systems generated by OpenFF and openmmforcefields workflows

  • Feels kinda clunky now, since wrappers don’t really provide functionality on top of wrapped classes

  • Uses some amount of





  • Should subclasses be allowed in validators/inputs?

    • Both: Yes

  • Validators/dependency definitions aren’t working. Could be a handful of reasons why. Agree that the idea is each component must defined its allowed inputs and some sort of error should be raised if something outside of that set it passed in

Action items

@Matt Thompson will continue working on
Look at old benchmarking code to crib energy minimization, energy evaluation, geometry comparison, etc. codes
Including passing conformers to these functions as positions
Add in some logic around molecule conformers when processing molecule inputs
Consider adding some openmm.app.PDBFile component in the topology module
Might be better to implement openmm.app.Topology, but they’re mostly interchangeable
Unclear current utility of either, but probably needs to be in place to use OpenMM’s protein force fields
Try to get a minimal, <100 molecule benchmarking run completed
Use some existing QCArchive data to bypass the need to do QC compute locally
Might be useful to use QCSubmit to do this, since that’s the choke point through which OpenFF will update its QCArchive interaction when the QC* stack breaks API in late 2022 or early 2023

Decisions