2023-08-15 OpenFF QCA Working Group Meeting Notes

JW – QCSubmit update:

• New build of qcsubmit 0.4.1 downpins to OFFTK <=0.14.0. So vanilla conda installs should work again.

• Main branch of QCSubmit is now compatible with new psi4 on conda forge (Thanks Brent!!)

• Expecting to fix OFFTK>=0.14.1 compatibility in next few hours (Thanks Brent!!)

• Expecting to fix industry dataset download in next few hours (Thanks Josh!) - planning to add warnings when invalid CMILES are read. Entries with invalid CMILES will be skipped when using OE backend. Will update RDK backend to skip them as well, though that will need to happen in the OFF Toolkit. Until both OE and RDK backends are updated, downloading the industry dataset will yield different numbers of molecules. This bug affects ~tens of optimizations of the 70,000 in the industry dataset. I’ll open an issue to discuss creating a new version of the dataset.

  • DD – QCS is OFF’s interface to QCA, so it makes sense that the fix would go in OFF’s stuff. Do we have a serialized copy of it anywhere connected to a publication?

  • JW – Not sure where to find it, but I’d assume “Yes”

  • DD – Ok, I’ll follow up with PB.

  • JW will open an issue on qca-dataset-submission to record this issue

• Python>3.10 compatibility is blocked-ish by a testing fixture with an import that requires python <=3.9 in QCFractal. Should we proceed without it?

  • …

  • Decision – We’ll stick with QCSubmit requiring python 3.9

• After this batch, I’ll make the QCSubmit 0.5.0 release+build the conda package

• After that, we’ll work on pydantic>=2 compatibility, and then update for QCFractal next

  • BP – Updating for pydantic2 is taking a lot of effort from our side.

  • JW – Can you do the import guard in the meantime?

  • BP – Unsure. I’m not sure whether objects using v2 and v1 legacy API can work together.

  • JW – If pydantic v1 and v2 objects don’t work together/compose, then things are very bad in our whole ecosystem.

• DD – So PE, to add fixes for SPICE 1.3, the shortest approach would be to use QCSubmit.

• MolSSI QCArchive user group

  • user questions / issues / feature request

    • LW – No issues here, I have serialized versions of everything I need.

  • BP – I’ve added the option to add new calculations “always”, instead of deduplicating. (kwarg is add_x(find_existing=True)and also dataset.submit). Still can't add duplicate molecules, but can add duplicate calcs.

    • DD – Awesome. I’ll push to add this to QCSubmit with the next branch updates.

  • BP – Working on metrics now. NSF site review next week.

    • DD – Happy to support if you need.

  • server instance statuses

    • QCArchive Legacy

      • v0.15.8.1

    • QCArchive OpenFF

      • v0.50.0b13

      • currently retains everything from Legacy

      • will lose ML datasets at some point

    • QCArchive ML

      • v0.50.0b13

      • mostly only contains ML datasets

    • QCArchive Validation

      • v0.50.0b13

      • MolSSI internal projects

    • QCArchive Demo

      • test instance

  • compute resources statuses

  • call for new users

• New datasets

  • qca-dataset-submission:

    • OpenFF Optimization Diverse Fragments with Iodine (w/ ESPs)

    • OpenFF Optimization Diverse Fragments with Bromine (w/ ESPs)

      • LW – Digging into both of these further, we definitely want more training data for Br and I, so these will be coming.

    • OpenFF Optimization Hypervalent Sulfurs (w/ ESPs)

    • OpenFF DNA

    • JW – Chapin has some datasets that he computed locally that he’ll want added to QCA when it’s ready for new submissions again.

  • spice-dataset:

    • SPICE 2.0

      • PE – Ran some tests, looks like we can use psi4 1.8.1. I’m just working on instructions for how to set up a manager.

      • DD – Gotcha - I’ll work with BP to get these to you.

      • DD – Also, running SPICE 1.3, do we want to use Psi4 1.6 or 1.8?

        • PE – Either should be fine

    • qca-dataset-submission-like automation on SPICE repo for handling execution?

      • PE – That sounds great to me.

      • JW – Who would own that repo?

      • PE – OpenMM / I would

• Resourcing updates from stakeholders

  • OpenFF

    • JW – Would be good to have more personnel-time for QCSubmit maintenance

    • DD – True, if we get more users we’ll have more avenues for resources.

    • DD – BP, will the new QCF be pretty stable?

    • BP – Yes, not expecting major changes.

    • DD – Right, so that should reduce resource needs.

  • Genentech

  • Chodera Lab / ASAP Discovery

  • MolSSI

    • DD – BP, changes to resourcing at MolSSI?

    • BP – We’re actually losing the postdoc who worked on the ML stuff. That impacts the exporting/importing stuff, and converting the existing datasets into more standard forms. I think I want to shift that over to a software scientist.

    • DD – Would that impact SPICE?

    • BP – Not in the short- or medium-term.

    • JW – Happy to talk to our leadership about getting MolSSI to allocate you more resources.

    • BP – Gotcha. I’ll be talking to DCrawford when he returns.

• QCFractal development

  • QCFractal v0.50.0 - imminent

    • milestone for completion?

      • BP – Planning to merge into main branch by the end of the month.

        • DD – That’s excellent, will probably look good at your site review.

      • DD – Want help getting started with public-facing milestoning/issue prioritization.

        • BP – I use a private asana right now, that helps me keep track in the time I have.

    • milestone for follow-up release?

• Additional business

  • MolSSI QCArchive Working Group start date: 8/29

  • BP – Yes. Will make a new calendar invite, figure out notes, make a new Zoom.