2023-01-11 Thompson/Wagner Check-in meeting notes

General updates

• JW – Monday is a holiday

• JW – Forked forcebalance, name may change but here’s the repo we can use: https://github.com/openforcefield/openff-forcebalance

• MT – I have no tasks related to exscientia stuff, ball is in their court.

Trello

Issues/PRs

• MT – Protein workflow speedups

  • Molecule.to_topology is slow and used in label_molecules, involved molecule serialziation roundtrips

    • JW – I’d be excited for a faster deepcopy, the serialization rountrip is just lazy and safe.

  • Pint unit deserialization takes 60% of protein ff loading time (1 sec of 1.5-2 sec). I don’t think this can be improved

  • ParameterAttributes have performance problems. For example, multiple k values are required to have the same unit when stored in a list. This added 100ms in loading time.

    • JW – If the speedup were greater this might be worth it, but 100ms one time for FF loading in this case isn’t too bad.

  • I may look into streamlining the chemical_environment_matches codes.

    • JW – This is dangerous and even I don’t really understand all the details of smarts matching, so maybe leave this alone and I’ll come back to it later.

  • There’s probably a lot of stuff I could improve in interchange.

  • I don’t see any evidence that things are so bad that users will have serious problems. T4 lysozyme takes 2-4 sec to load into a molecule, total 15 secs to make an openmm system. I’d like to cut that down a lot.

  •  

• JW – T4 is a little small, maybe check for super-linear scaling in something bigger like https://github.com/openforcefield/protein-ligand-benchmark/blob/main/data/pfkfb3/01_protein/crd/protein.pdb (400-some residues)

• MT – Can I just make a protein from sequence?

  • https://www.rdkit.org/docs/source/rdkit.Chem.rdmolfiles.html?highlight=molfromsequence#rdkit.Chem.rdmolfiles.MolFromSequence

  •  

  • 
    
    

openff-forcebalance plans

• MT – How to version fork? Continue existing system? Start at 0?

  • JW – start with 0.0.1

• Repo metadata cleanup

  • Update README to say “this is a fork of LPW’s forcebalance, you’re probably interested in using that instead. For that code, install instructions, published parameters, citation information, and funding acknowledgements, follow this link”

  • License under MIT

• Repo unused feature cleanup

  • Remove GUI, Tinker interface, other out of scope functionality

• Repo structure updates

  • Update package structure to use openff namespace

  • Remove cruft from setup.py

  • Remove Python 2 code

  • Skip all failing tests

• Set up new automation

  • Update CI

  • Set up CodeCov

  • Version with versioneer.py

  • Set up pre-commit and pre-commit.ci

• Code cleanup

  • Add some type annotations

  • Un-skip previously-failing tests

• Iterate on expanding unit tests

  • See what code is not covered by unit tests

  • Add unit tests

  • Run unit tests

• Iterate on expanding examples

  • Add back/curate examples

  • Run examples

  • See what code is not covered by unit tests or examples

• Refactor BespokeFit to use fork

• Update Evaluator interface

• Update Recharge interface