2024-04-03 OpenFF/FE collab

Owned by Richard Gowers, created
Last updated: Apr 08, 2024 by Matt Thompson
2 min read

Participants

  • @Matt Thompson

  • @Richard Gowers

Goals

  • establish current status of AFE workflow functionality

  • figure out next steps for progressing project

  • figure out new deadline

Discussion topics

Item

Notes

Item

Notes

 

  • new solvation code path seems to work and produce sane reported results. need to run this for meaningful amount of time with either code path to check for estimate equality

  • analytical System equality check

    • fuzziness allowed (all due to solvation):

      • sigma on hydrogens (these are 0)

      • system volume dimensions

      • number of waters

      • position of waters

    • should be strictly equal

      • barostat

      • thermostat

      • small molecule forces

      • FF applied to water

      • constraint distances

      • masses

  • confidence in existing packmol implementation is high, this is tested elsewhere

  • initial packmol configurations might be a bit sketchy

    • RG: Can adjust the initial equilibration parameters

  • full scale validation on HPC is openff responsibility

Initial Validation requirements?

  • for 5 ligands in water, statistically identical results with existing and new code paths

  • pending getting SFE data

    • cherry pick 10 “well behaved” small molecules in non-aqueous solvent

    • acceptable agreement with experiment for 10 ligands in non-water solvent

  • this should be computable in ~week

Future steps for solvation workflow

  • will probably “release” skunkworks as a tag on github

    • MT to check if this is sufficient for initial openff requirements

  • will eventually fold this work back into main openfe release line once we’re happy with final product

    • RG responsibility to finally merge this into main openfe repo

  • after merge openff team can use mainline SFE openfe repo for this work

Action items

@Richard Gowers to provide script to produce System/XML representation of prepared system post alchemy
@Richard Gowers to pick better (large enough to feature torsions / nb exclusions, w/ more elements) molecule than benzene from freesolv dataset
@Matt Thompson to take this and write fuzzy equality (defined above) function to check forces
@Matt Thompson will add some unit tests in Interchange to make sure ligand-in-non-water-solvent PACKMOL code paths work as anticipated
@Richard Gowers to check benzene hydration with both code paths for equality before HPC stress test
@Richard Gowers to create script for HPC validation by openff personnel
@Matt Thompson will talk with science team about SFE data, plans for SFE in fitting and/or validation
@Richard Gowers to agree on initial validation set
@Matt Thompson to co-ordinate running of initial validation set
@Matt Thompson will get confirmation from science team that working on on github tag (no conda package) is okay for production-scale SFE benchmarks

Decisions