2023-08-08 OpenFF QCA Working Group Meeting Notes

PE : questions on submission, execution in the context of SPICE

• BP : openff-qcsubmit

• DD – SPICE 1 was submitted and managed using qca-dataset-submission repo. openFF was willing to take on the dataset composition and submission for that, since we thought it would be simple. It ended up being more challenging than anticipated. Since SPICE isn’t an OpenFF project, OpenFF declined to do much more past SPICE 1.x.

• JW – recommended way to proceed depends on what you need to keep track of

  • if you’re starting from graph molecules, openff-qcsubmit makes sense to start from

  • however, openff-qcsubmit is currently broken (does not support new PortalClient API)

  • PE – current recommendation is to use QCPortal directly?

    • JW – at the moment, yes; assuming you don’t have graph molecules

    • PE – what’s a graph mol?

    • JW – kekule structure

• PE – input from HDF5 files was parsed using what code?

  • DD – that implementation lives in openff-qcsubmit

  • PE – so can’t be used at the moment?

  • DD – not at this moment; potentially in a few weeks once openff-qcsubmitis compatible with latest QCFractal

• BP – Contents of HDF5?

  • PE – Coordinates and mapped smiles

  • BP – That could probably be brought into its own script.

PE – how do we specify spin multiplicity for each molecule in Psi4? This information is not included in the input HDF5 file we used

• BP – I thought these were all closed-shell

• PE – I think spin multiplicity still needs to be specified.

• PB – I think we chose the default one if neutral, if charged… 2*S+1

• PE – I don’t know of a simple rule that would give a default multiplicity

• BP – It assumes closed shell singlet. That should work unless there’s an odd number of electrons.

PE – Differnet ways of rerunning the bad calcs when there was a bad version of psi4

• DD – What we would have done is created a 1.(N+1), where the entries that we want to recalculate are translated by 1 bohr, and leaving the others alone. So the the new dataset would have all the molecules, but only the mols that we want recalculated would be rerun. Is this still the way we’d want to do it today, BP?

• BP – In the new version, records can be deleted. But if you want to keep the old version around, I’m planning to implement a feature that will let you submit duplicates. But before that happens you can still do the shift-by-one-bohr trick.

PE – How can I pin to use a specific version of psi4?

• BP – There’s no way to pin to use a specific version of psi4, but you can do it via the tag system. A manager will only pull down records with tags that it accepts. So you can run managers with a specific version of psi4, and then have them only accept jobs with the same tag.

• PE – So I can tag all my jobs with a psi4-1.6.1 tag, and only the appropriate managers would pull down those jobs.

• DD – Yes, but the tags are all human-set, so the person who starts up the managers would need to ensure the right psi4 is installed and the tag is set.

• PE – Which version to use? I’ve tested 1.6.1 but there are newer ones.

• DD + BP + PE – …

• DD – So we should rerun these tests on newer versions (1.7, 1.8) before we submit the whole dataset.

PE – Manager setup instructions?

• BP – Working on updating those. The old setup docs may work to some extent but there are some differences. Also there are enough different possible cluster setups that a one-stop-shop compute manager guide will never be complete.

PE – New QCF code - Is it ready for use, or should we hold off until it’s finalized?

• BP – Ready to use. Planning on no breaking changes for a while. Will post new docs and install instructions.

• New docs at https://molssi.github.io/QCFractal/index.html

• New packages at qcarchive anaconda label https://anaconda.org/qcarchive/

• DD – Could we move the changes in next to the main branch, and unlabel the conda packages?

• BP – Not sure, very busy this month, not before september.

JW – forwarded contact from quantum computing group at IBM

In progress datasets

• DD – None

Compute resources status

• DD – None, but working on getting Lilac running.

DD – New people who want to run compute managers?

• CI – Looking at some active learning stuff and incorporating into our code.

• NF – I’m interested in joining with SPICE 2.0 from Genentech’s side.

• DD – PE, could be good to collect interested parties with SPICE2

DD – Training session videos are up and linked on the #qcfractal channel.

DD – Server software status/versions?

• BP –

CI – With the new branch that’s being developed, are the old things languishing/if we submit issues will things be done?

• BP – To the next branch, yes. To the master branch, no. I know there was some trouble with the pydantic version bump.

• JW – MThompson and JRGuerra retroactively pinned old QCPortal packages, but any future packages will need their recipe pinned.

CI – Should I be buidling against new branches? If so, how?

• BP – Yes. I’ll send docs.

• JW – Yes, build against new versions.

  • New docs at https://molssi.github.io/QCFractal/index.html

  • New packages at qcarchive anaconda label https://anaconda.org/qcarchive/

DD – SPICE dataset discussion?

• PE – SPICE is intended to be growing with time. So SPICE 2.0 will explore more chemical space, things like protein-ligand interactions, ligand-water, include coverage of boron and silicon.

• DD – There was a talk at SciPy about other folks who used SPICE in their own work.

• NF – Myself and other scientists were planning a QC dataset for training potentials, and when we found Peter’s paper it was exactly what we needed. So we’re interested in joining forces. So what we talked about in the SPICE2 meeting was flagging the interesting datasets and providing compute.

• DD – One of the challenges was that SPICE2 doesn’t have a home - there are resource contributions but no core driver yet.

• PE – Kinda. We have resources at Stanford that we can provide for doing this. We should have more nodes this time around than last time. Everything is managed through the SPICE dataset GitHub repo.

• DD – What’s between us and submitting SPICE2?

• PE – Manager documentation should be the big one, after that I can get it running.

LW – OpenFF datasets - These are hung up on QCSubmit but the three listed in the agenda are top priority. Would like to have that submitted in the next month, and have results in the next month or two. Highest priority dataset is Hypervalent Sulfurs.

• JW – I think we can do that

• PB – Update on bromine and iodine?

• LW – Sulfurs are highest priority since we don’t have any of those, I think bromine and iodine would come after that.

JW – openff-qcsubmit

• three major issues blocking update

  • QCF API change

  • Removal of ChemicalEnvironment from OFFTK

  • Problematic molecules in existing datasets (like industry benchmark set)

• I’ll fix 2 and 3 first, then will work on 1.