2021-07-26 Core Developers meeting notes

Updates

• SB

  • Nothing too major, working through bespokefit. Trying to get the backend into a more stable, restartable state

  • Working on recharge wrt Trevor’s recent additions

  • Working on QCEngine torsiondrive PR / additional testing

• CC

  • Focusing on getting a dipeptide torsiondrive set done. Focusing on good starting conformations – Considering using the conformations from the amber-ff-porting. Are there other options?

    • JW – Could use a current openff forcefield and use torsiondrive/geomeTRIC to generate starting coordinates

    • CC – To clarify, my question is “do we have a tool to rotate a dihedral toa desired angle ang generate a conformer with that?”

    • SB – GeomeTRIC could do this – You’d add a constraint for the dihedral of interest, and then it could minimize.

    • CC – JW had said to generate many conformers and select the one with the lowest electrostatics, but I’d think that low sterics would be better

    • SB – In principle, you can give torsiondrive many starting conformers, and it will be able to start the scans from the closest initial conformer to each of those.

    • DD – That’s correct.

    • JW + CC – The goal here is to be able to force inclusion of sidechain rotamers of interest in the starting set.

    • SB – Yes, so this would probably look like 1) generate a bunch of unconstrained conformers of the dipeptide of interest 2) constrain/select the sidechain angles of interest 3) dispatch torsiondrives from the desired starting positions

    • CC – Which openFF forcefield should I start with? I’d like to be consistent with which FF I use moving forward, both for conformer generation and as a starting point for training the biopolymer FFs.

    • SB – It doesn’t really matter. I think Sage RC2/Sage full release would be a good starting point.

    • CC – Anything else I should have ready before we have a working session on this submission?

    • DD – Nope, I’ll plan to meet with you and have a working session.

• MT

  • Mostly worked on vsites in Interchange, coming along pretty well. Have some simple cases working in OpenMM and GMX. Working toward comparison, though some of the position geometry/math is tricky. IT looks like vsite water models will be tricky, since SMIRNOFF decouples bond/angle constraints from vsite geometry whereas GMX has vsite forms that assume certain bond lengths and angle values (sometimes a vsite’s distance parameter is just a unitless scalar times a bond length).

  • Decent amount of Toolkit work.

    • Some AMBER infrastructure/ParmEd debugging.

    • Found that, since we don’t have all exceptions consolidated into a single file, we have two ParseError classes defined in different places. I’ve opened a PR to resolve this - If we fix this quickly then the conflict will never appear in a release.

      • JW will review this before 0.10.0 release

  • Testing OpenMM 7.6 beta

    • Big change is that import paths changed (from simtk.openmm import ... to from openmm import …. Eastman made most imports work either way, but a few deep imports don’t follow this patters. I found one deep in Topology.to_openmm and opened a PR to fix.

      • JW will review this before 0.10.0 release

    • MT – Not sure if we should report this upstream, may not be worth the effort

  • All of ParmEd’s openMM module is going to break with the ParmEd 7.6 release. Opened issues and PRs to resolve.

    • It’s actually fixed in Parmed:master, but there hasn’t been a release with it yet, and I’m not sure whether the next AmberTools package will pull from GH or just not be updated.

• DD

  • Short week last week

  • qca-dataset-submission: Got a set from Pavan submitted, moving pretty fast

  • Trying to get MM industry benchmarking sets submitted.

    • There were technical issues regarding submission, now I’m not sure whether they’re actually running or not. So I’ll be doing some checks on the compute managers

    • I’ve started a MM compute dataset starting from the completed QM jobs. I’ve generated the dataset jsons locally and will submit them soon.

  • Prepared modified datasets using the alternate basis/method that should handle iodine better.

  • Working on GH action to test psi4 release candidates/newer versions

  • Benchmarking

    • Worked with LD to get new Swope/Lucas analysis merged in. This was a good exercise and LD got very confident about using GH/opening PRs

    • Worked with JW on openff-benchmark and sage release candidate testing

    • Worked with Xavier Lucas on torsiondrive executor. He can run them now, but needs help parsing molecules from json.

  • PLBenchmarks on F@H

    • Preparing first draft of proposal to gather feedback from stakeholders. Expect to do one more revision after this, and then move on to prototyping.

• JW

  • Benchmark stuff/sage rc1 debugging

  • RDKit conf gen bug, will post shortly.

  • ParmEd silliness / 1-4 terms debugging

  • Custom OBC GBSA ParameterHander with Setiadi

  • ISMB talk 4 AM on wednesday, may be less available Tuesday as I prepare talk, may take off early on Weds if I need sleep

Sage release?

• JW – I’d like to send the partners some form of Sage to benchmark this week. Not sure if it should be Sage “RC2” or Sage “full release”.

• SB – Need to think about it more. Not sure whether there’s any value to RC2, since the past work hasn’t found anything. I’ll make a call later this week.

• JW – If we’re still thinking on Wednesday, I’ll push to just make an RC2, and then we can keep thinking about it.

Toolkit 0.10.0

• SB – Would like this release to include #940

  • MT – I’ll reread the discussion on #940 and identify and final blockers

• JW – I’ll aim to merge remaining PRs (such as charge normalization) ASAP.