2024-05-08 Meeting notes

Filtering

MT

• MT – Some loose end about why alkyne pattern hits different stuff than checkmol patterns. We could delve in but I don’t think it’s super significant.

• LM – LGTM

Atom ordering bug

LM

• LW – Will there be a problem if two “molecules” are added with different mapped smileses but identical inchis?

• MT – In conformer array, each row is indexed by its atom ordering. I can’t say confidently though that there are safeguards in place to keep that bad state from happening. I should check this.

• LW – May start happening as we start assembling/combining datasets. So should be a thing we check for the future.

• MT – Good call, will add to trello.

Torsion datasets

MT

Good torsion dataset to pull from? Might filter out a subset of molecules for scale convenience. Right now I think we’re just storing OptimizationDatasets, but now I’ll need to start storing TorsionDriveDatasets. Should I expect to have Optimization and TorsionDrive data in the same store?

• LW – I don’t think so. You can’t really do unconstrained opts wiht torsiondrives, since TDs need to be constrained/don’t like that.

• LW – For dataset choice, the most recent supplement torsiondrive dataset is reasonably small. BW, are there any where the torsiondrive hasn’t finished a full scan?

  • BW (chat) – I was just going to say that I have filtered my second-to-last torsion drive set for charge issues, so I could send you that. I have not looked at whether the scans finished, to answer Lily's question.

  • https://github.com/openforcefield/qca-dataset-submission/pull/354

  • LW – I’d be careful with those - Check for ring torsions that didn’t finish rotating.

  • LW – Protein torsiondrives would be good as well, since they’re 2D as well.

General discussion