2020-10-05 Core Devs Coffee Meeting notes

Roundtable Updates

• DH – Not much new. Preparing manuscript.

• MT – Nothing huge.

  • Progressed some PRs.

  • Applied to BSSw fellowship. Skeptical about odds of getting the fellowship.

  • Met with Vanderbilt on interoperability. Interop meeting may have gone long, could have planned ahead and cut this off/kept it more focused. I think we should focus in the short term on converters, and that will “scout” our the feasibility of bth using a single object. For the next two weeks, we’ll spec out what our system objects need to do (from the perpective of behavior, not internal structure), and then we’ll compare our final specs with Vanderbilt’s. Worried that I’ll get paralyzed by speccing too much and not implementing.

  • WRT Yutong’s timemachine talk -- new System will be able to appear in an auto-diffable form

  • DD did a thorough review of WBO bond PR.

  • JW did a review of CLI conf gen PR.

    • JW – Complexity regarding reading multiple confs/multiple molecules. Maybe future CLI tools could follow a “one molecule/one confomer input” philosophy

    • MT – Agree, we should make a table of expected behaviors WRT multiple molecules, multiple confs, multiple charge sets. Maybe a blog post

    • JW – I could add this to docs.

    • MT – Maybe also long sections in release notes could be added to docs.

  • 0.7.3 release this or next week? Would let us get N-1 CIMH PR.

• DD –

  • Reviewed WBO bonds

  • Lots of work on QCA. Worked with BP on new release of QCF. Better handling of INCOMPLETES

  • New QCE release in conjunction with Lori. Performance improvement in TorchANI harness.

  • New envs already deployed. Ran PRP over the weekend. Will send update for manager operators.

  • Made value stream diagram, folded in System/Evaluator aspirational connections, and bespoke workflow.

    • Would like feedback on current diagram – Will be one or two more iterations.

    • Link: https://drive.google.com/file/d/1LM7sBI446YHY_GepRgUxPXTDC4Uz9HSC/view

  • Working with DH and GT on benchmarking, will be joining further calls with industry benchmarking efforts.

    • JW – DH messaged us that Janssen wouldn’t be interested in working with Galileo. Would other companies possibly be interested?

    • DH – I doubt that other companies would be interested but we can bring it up.

    • DD – I think it shouldn’t be plan A.

    • DH – We get the feeling that setting up QCF workers will be easier than doing the paperwork/legal stuff to get galileo access.

    • DH – Collaborating on FAQs/tutorials seems like far less work to us than Galileo paperwork

    • DD – It will be somewhat complex to get QCF workers set up on pharma clusters with different topologies. I look forward to hearing about the rough scale of what partners want to do, and to see whether it would take a few weeks or a few years ot run.

    • DH – GT mentioned ~100-1000 molecules, 10 conformers each → ~1,000-10,000 optimizations. Janssen would like to run a test over Christmas holiday, with other pharma partners running later.

    • DD – We’ll aim for this effort to give us information about what our ultimate benchmarking infra will need to do, but not necessarily plan to have this seed our ultimate benchmarking infrastructure. We’re starting with optimizations (not torsiondrives) to keep things simple.

    • JW – Reviewing conformer generation CLI.

    • DD – What file format do you anticipate using?

    • DH – 2D SDF / 3D SDF / SMILES

    • JW – Maybe we could ensure stereo is fully defined by setting allow_undefined_stereo=False in OFF Toolkit molecule creation methods.

    • DD – Will try to fill this in when we meet with pharme partners.

  • Starting to work with SB on Evaluator.

• JW –

  • 0.7.2 release. Automation for SFIs saves HOURS and lowers release friction a lot.

    • Releasenotes polishing and RST → MD conversion is still a pain.

    • Would ike the conda workflow to be able to point to an omnia/label/rc file, so tha tI don’t push the new package directly to omnia/label/main to test the new package on GHA.

      • JW – I’d be in favor of nightly builds, to be kinder to our downstream users.

      • DD – Do they even run nightly tests? Would want to add this to docs and developers guide.

      • MT – I wasn’t in favor of nightlies. It won’t be hard to build, but until there’s a clear need, we shouldn’t invest in this. Let’s establish that there would be people who would use this before we chew up CI time / Anaconda storage space.

      • JW – I’ll figure out how to query likely consumers about this – Could post to infrastructure or ask a PI to tweet.

      • DD – Yeah, let’s ask directly about this and see who’s interested.

    • Built Fragmenter 0.0.7 package, omnia only had up to 0.0.4, OETK 2020 dependency breaks this. Waiting for Horton to test before moving from RC to main

      • DD – It’s become clear that industry benchmarking will require fragmenter refactor to become OE-independent.

      • JW – I’d like to contribute to this.

      • (General) – We will open up a branch of Fragmenter to work on this. DD and JW will open and review PRs into this branch. Will decide timeline at next sprint planning.

      • MT – Let’s make sure we keep previous lessons-learned in mind as we move forward on this.

    • Conda commercial use panic – Spoke with Quansight and they think we’re pretty safe, but are looking into it. We basically want to pay “our share” and not cover, eg, all of Pfizer’s usage.

      • Nuclear option – Package our own stuff, host our packages on Quetz running on OpenFF server.

    • BSSW fellowship application.

    • Working through PR reviews/issues.

      • MT – Do N-1 CIMH first

    • Didn't get to N-1 CIMH

• PB –

  • Pushed update to Topology.to_file where we print PDB representations of topology.

  • Started looking into IUPAC exception. Didn’t quite understand what difference was seen. When I looked into the tests, one uses minidrugbank.sdf, other is tripos mol2.

    • MT – Could have been clearer about the observed difference here. Didn’t expect you to get to this so soon. I’ll write up a more thorough summary of what’s happening here.

  • Still need to get to MDAnalysis / from_pdb functionality.

    • JW – Let’s talk before you get started on this. We’ll want to makea dataset of structures.

  • Working on WBO analysis/barrier heights.