2022-02-28 Core Developers meeting notes

General updates

• Bug in Toolkit since 0.10.1 would have caused waters parameterized by Sage to get AM1BCC charges

  • This change was introduced in #1036, which was merged Aug 13, shortly before the Sage release (Aug 16 2021). The change didn't make it into a stable release until toolkit version 0.10.1, on October 26.

    So the Sage fitting should be unaffected. But users using OFFTK 0.10.1 or 0.10.2 would have been getting AM1BCC water when they should have been getting TIP3P.

• PB – Next week is CUP - Myself, Mobley, others will be partially offline Mon-Fri.

  • JW – I’m not planning on going unless specifically requested, I can’t justify adding time to the toolkit biopolymer release.

Individual updates

• CC

  • Out most of last week for Biophysical Society Annual Meeting

    • Poster on Zenodo here: https://zenodo.org/record/6153446

    • 2022-02-24 Protein FF meeting note

  • Putting more time into LiveCOMS review, trying to get that out to external authors

  • Need to check in on Protein torsiondrives, need to check to see whether we have the same problem with resubmitted arginine

  • PB – Mind if I resue parts of poster for CUP poster next week?

    • CC – Sure. I did that poster in Latex, can send you the raw files.

• DN

  • Preparing a document to better understand our strategic planning. I have to answer a lot of questions, and maybe I will come to you for some information.

    • DN – Not just for OpenFF, instead for OMSF as a whole. I’m trying to put all tasks in the same “bucket”. So we can put everything in a larger context to identify the value and development priorities.

    • Both OFF and OFE will have projects. I can provide more details later. I’d like to use this for my thesis if there are no objections.

    • JW – I think that’d be great - It’ll be good to have the ability to coordinate OFE and OFF’s work so that we put things out in an order that will provide value to the outside world.

• MT

  • Opened OFF-EP 0005

    • Splits the electrostatics method into two - one for when a force field is a periodic system and another for non-periodic system

    • Anticipate no impact on 99% of use cases

      • expect toolkit to automatically up-convert the common case of PME/no-cutoff

      • copy-pasted <Electrostatics> header would be updated

  • Finished first iteration of condensed phase regression tests

    • Fixed one quirk with constraints that should not impact behavior - everything else matched in openmm.System XML files to 8 sig figs

  • Value propagation regression tests

  • Tinkered with NAGL for assigning charges via graphs

  • Conda package for bespokefit is online

    • All I did was package - Josh H, Josh M, Simon built the thing

  • Submitted an abstract for a talk at SciPy (second week of July in Austin)

    • Some parts of our infra lack branding/public-facing things like citation mechanism

    • JW – I’ve tasked JM with doing some videos this year showcasing what our infrastructure can do - If you’re interested in helping improve visibility, it may help to give him feedback.

    • MT – Last week I got asked whether there’s a citation for interchange. I realized that there isn’t. This sort of thing will never be urgent but it will always be important.

    • JW – I’d be interested in a review of “how should we add DOIs to papers?”. Right now OFFTK has like a bazillion zenodo DOIs, one for each release.

    • MT – Maybe take a look at GH citation machinery?

    • CC – Maybe only have new DOIs for significant version bumps?

• PB

  • Spent a day on espaloma benchmarks.

    • Tested on industry benchmark set. This was a broader test. Earlier YQ had found that espaloma did better than OFF on mols with bridgehead nitrogens, and this spurred DM asking me to do a mroe general benchmark. I had trouble getting espaloma set up, the install instructions didn’t work out of the box. My theory is that, since espaloma was trained on optimized geometries, it doesn’t capture some of the torsion info.

      • PB – Specific install things that could be documented better:

        • Needed to update the dgl-cuda version to be appropriate for my cluster

        • Needed to pip install pytorch

      • MT – I’ve run into similar issues before. Conda packages in the ML space are a bit unwieldy. And espaloma is outside the OpenFF umbrella.

      • DD – Do they have install instructions anywhere? We could raise an issue.

        • PB – I initially used https://espaloma.wangyq.net/install.html

        • PB – Yeah, I DM’ed YQ and he instructed me to use conda-forge and dglteam.

        • DD – It’d be good to raise an issue about the broken install instructions, to motivate an update.

    • JW – Was this espaloma assigning parameters to the molecule, or espaloma calculating the entire conformer energy?

    • PB – I think it just assigns conformers, not sure.

  • Spent rest of the week on fitting experiments.

• JW

  • Will make 0.10.3 release today, see above. Will likely no longer accept PRs to master.

  • Working on vsite tests

  • Looking for bespoke beta testers (Thanks for help with the release JH, SB, MT!)

  • F@H planning

    • JW – CC, could you join F@H planning meeting tomorrow in case we have further questions about protein observables?

    • CC – Yes

• DD

  • Protein-Ligand Benchmarks

    • had our second working group meeting, reviewed user stories on fah-alchemy with story submitters

    • I'm evaluating these stories against the architecture, and working to lay out initial set of issues for MVP project board

  • QCArchive

    • migrated qca-dataset-submission to use git-lfs; reduced repo size to 108MiB, so clones are now very fast

      • new *.bz2, *.zip, *.gz files committed by users are added to LFS automatically

      • Any existing checkouts will need to be re-cloned

    • compute deployments on PRP and Lilac focused on ESPs, protein capped 1-mer sidechain torsiondrives

      • request from Chodera to make SPICE priority on Lilac

      • PB – It’s fine to reprioritize that. Thanks for coordinating.