2022-11-03 Meeting notes

QCF difficulties

• MT – It’s really hard to query things on QCF, so it’s hard to collect even an initial test set of molecules. I’m going to work for now assuming that we start with a direcotry of SDFs similar to benchmarking season 1.

  • JW – That sounds good. QCFractal next is expected around Dec, though given the MolSSI workload there’s a good chance it’ll be late.

QCEngine difficulties

• MT – In the course of trying to get things to work, DD showed me a place where QCEngine calls geometric calls QCEngine, which is not great. It looks like BPritchard/MolSSI isn’t really maintaining it. But LBurns is making a new release now.

• JW – Relatedly, next year we may start hosting our own QCF server, separate from QCA (the lion’s share of MolSSIs hosting costs right now are OpenFF). So they’ll keep developing the software, but we’d be paying to host our own instance.

WBOs in SDFs?

• LW – Was wondering about whether we could serialize WBOs into SDF.

  • MT – I assume this would go into an SD tag, like JC mentioned?

  • LW – It looks like RDKit doesn’t read SD information on to bonds.

  • JW – I think the crux of the possible problem is that OFFMols created from SDF are guaranteed to have the same atom order/index as the file, but that same guarantee is not made for the order/indexing of bonds. So if we had an SD tag like bond.dprop.fractional_bond_order and RDKit didn’t put it in each bond’s property dictionary, then we might have a lot of trouble matching them up later.

  • LW – Also, partial charges and fractional bond orders aren’t sent to qcschema. Nagl users may want an easy, common file format (e.g. SDF or JSON) in the future.

  • JW – Right, I don’t think we guarantee that either goes to QCSchema currently, but we can add it if it’s a need for the fitting team.

  •  

Distractions

• MT – QCF example in toolkit not running?

  • (General) – Upon review of most recent CI, it looks like it is running. So this was probably a past issue.

• JW – Throwing a vsite ff over the wall to Exscientia ? At worst they do nothing, at best we get a better spec for AMBER vsites, or they write our exporter for us

  • MT – I kinda got the impression than “exscientia writes the exporter” isn’t in the cards. They came with the attitude that I’d write it and they’d test it. So I didn’t get the sense that they’d give us “free work”.

  • MT – I found a lead on actual ways to add vsites to prmtops: https://ambermd.org/tutorials/advanced/tutorial35/index.php

  • MT – It’d probably take me a few hours to get their engineers up to speed on how to interface with Interchange internals. Most effective non-monetary way for them to contribute would be as testers. But overall I don’t think we need to up this in priority right now. Also, even if we DO implement this sooner, we still wont provide any production FFs with vsites any sooner, which I would imagine to be the primary benefit they’d expect.

  • JW – Agree. If they come back to you with an interest in trying more work, let’s respond with the offer to send them the vsite force field (and a lot of disclaimers). Please loop me into further emails/communications.

• MT – Lester Hodges was at exscientia meeting, was at OpenBioSim, seems like they’re trying to do something similar to OMSF.

•