2022-09-26 Core Developers meeting notes

General updates

• JW – Toolkit 0.11.1 released, cracking down on radicals. Seems to have broken Fragmenter, JW will look into this.

• JW – Follow-up workshops this week: Biopolymers+PTMs on Tuesday, Interchange on Wednesday.

• JW – Dotson is out today+tomorrow

• JW – May take half day on Friday

• DN – Should we move this meeting back yb 30 minutes?

  • (General) – Yes

  • (Meeting moved to be 30 minutes later)

Individual updates

• DN

  • Last week I updated website, corrected an ad board note about potentially embargoing new FFs (miscommunicated to include software)

  • Worked with OpenFE to make project page

  • Talked with Virginia Burger - She said that her company was talking to NVidia soon about GPU optimizations. Wants to join together to offer advice to nvidia.

    • JW – I don’t know mucha bout GPUs of what we could tell them, but it couldn’t hurt to start a dialogue

    • CC – Think about future functional forms - Polarizability/DEXP? They may be able to optimize for that.

• MT

  • openmmforcefields is updated (again) to 0.11.2, this time with tests that ensure current and backwards compatibility with the OpenFF stack. Let me know if there are issues.

  • Minor progress on units/models packages - reach out if interested

    • GitHub - openforcefield/openff-models: Helper classes for Pydantic compatibility in the OpenFF stack

  • Tried refactoring GROMACS export, unfinished

  • Workshop prep

• CC

  • Capped 3-mer backbones QC dataset has completed 12/54 TorsionDrives up from 6 last week

  • Found a problem with torsion assignment for delocalized charges due to SMIRKS strings in our FFs. Issue here.

    • CC – I’m making some new symmetric carboxylate parameters to correct for this asymmetry. Only applies to aspartate, since glutamate sidechain is too long and goes beyond the two sidechain torsions I’m fitting.

  • Protein parameter fits are progressing slowly. Working with DM on getting set up on UCI cluster.

    • Many ForceBalance workers die due to idle timeout. Consistently, only three workers run simultaneously. Is this a problem with my job setup? On my cluster, only 3 of my jobs actually get running, regardless of how many I submit.

      • PB – Does the output indicate that the workers can’t communicate with the head node? I sometimes see things with ports.

      • CC – Maybe, but the jobs just say “idle timeout” because they weren’t assigned any tasks

      • PB – Does JSetiadi run ForceBalance jobs on TSCC?

      • CC – He uses a smaller Gilson lab cluster, but we assess that it’s too small for the protein FF fits.

      • PB – Ok, then the top suspect is worker communication with the head node. There may also be an option where you increase the time limit… let me check what I did.

• JW –

  • Toolkit 0.11.1 release - Forbids loading radicals as OFFMols.

  • Helping with followup workshop prep

  • Preparing tea times

  • Good ad board meeting - Got OK on spending infrastructure time on non-openff software, software reliability

  • Led F@H meeting. Seems like PL dataset reformatting is going to miss target, but we don’t actually need it until December at the soonest.