2024-06-12 Meeting notes

 Date

Jun 12, 2024

 Participants

  • @Matt Thompson

  • @Lily Wang

  • @Alexandra McIsaac

  • @Brent Westbrook

  • @Jeffrey Wagner

 Discussion topics

Item

Notes

Item

Notes

Org-wide models for datasets?

  • https://github.com/openforcefield/yammbs/pull/51/files

  • MT – Would like to think about input/output format models. Not sure it makes sense for QCSubmit models to literally be the models that we use internally in YAMMBS. On the other hand, I see the downside to having our own models since those might proliferate into like a dozen models.

    • LW – I think you asked whether it’s worth it to have YAMMBS-specific models, or more efficient to have more general models that YAMMBS can use a subset of. I’m not sure I can evaluate this immediately, my off-the-cuff response is that it’s hard to have a really general model - We’ll keep discovering new use cases and need to extend/break the model. MolSSI kinda has these models and they’re not perfect but they have evolved a lot. I’m not sure against having our own general models since the flexibility is a big upside. But on the other hand I’m concerned about duplication of effort.

    • MT – I’m not advocating a broad community standard for molecule collections, would just scope it to OpenFF stuff, not setting out to make everyone happy. I wouldn’t set out to be super general, rather just have a list of QM reference molecules.Though I don’t strongly disagree with anything you said.

    • JW – Largely agree with LW.

      • I dont' think we have the domain knowledge to spin our own QC molecule class.

      • It would require compelling reasons to turn away from MolSSI’s open standards and spin up our own.

      • I’m not married to QCSubmit classes, so it owuld be fine if YAMMBS sidestepped those.

    • MT – I don’t think that we should spin up our own QC molecule objects. I’d see our objects as more of MM reference molecules build from QM molecule. It could be handy to have some history of the QC info that brought the MM reference molecules into being.

    • MT – It’d help me to understand what it is we need from QC land in our reference molecules. Coords, connectivity, energy?

      • LW – I think that’s about it for optimization datasets. For torsion data we’d need to know which torsions are constrained. Also grid reference points.

      • MT –

      • JW – Maybe YAMMBS mols could be a MolSSI molecule+some minimal stuff we need on top of it?

      • MT – Yeah, could either subclass from or contain MolSSI mols. Mostly interested today in figuring out whether there are greater standards coming together that I should aim to join with.

      • JW – I wouldn’t call this a standard, rather an implementation intended for our science team with no guarantee of stability or recommneded use outside.

      • MT – Maybe standard was the wrong word - Basically looking to figure out if there are greater plans like around FB targets that this would aim for compatibility with.

      • LW – Might be a large project to switch lots of our infra over to use that model. Might be best to make this a YAMMBS-specific model now and consider broader adoption when we’re doing future refactors. Eg, hard to plan for FB target replacement since FB mols would need a ton more information.

      • MT – Are you suggesitng basically developing this model just for YAMMBS now and then consider broader applications in the future as opportunities arise?

      • LW – Yeah, that’s basically what I’d advise.

      • Decision - keep moving forward with what’s started in

      •  

  •  

  •  

  •  

  •  

  •  

General discussion



Trello

https://trello.com/b/dzvFZnv4/infrastructure

 Action items

 Decisions