2021-03-17 Benchmarking for Industry Partners - Development Meeting notes

QCElemental charge and multiplicity

Lorenzo

• DD: checked to make sure charge, multiplicity passed from openff-toolkit Molecule to QCElemental Molecule in our benchmarking pipeline; confirmed

SLURM issues - memory

Lorenzo

• Several workstations connected as a workstation queue via SLURM; these are also user machines

  • DD: Janssen is using compute option (C) to my knowledge; doing conda env create on individual workstations as first part of compute script

  • if hitting memory issues, try setting memory fed to openff-benchmark optimize execute to a value much lower than you are requesting from SLURM. If you ask for e.g. 36GiB from SLURM, give the command 24GiB.

Release

Jeff

• Who should make the release today?

  • @David Dotson will lead the release; Jeff will join right after this call

SMARTS filter

Jeff

• Boron is non-fatal, but want to filter structures with iodine

  • this will be next release, following today’s

Schrodinger executor

Jeff

• Jeff in a position to review the Schrodinger PR from David H.

  • will give it a try, note issues on PR

Coverage filter for Boron

Josh

• Can’t reproduce the error; a boron-containing molecule doesn’t make it through coverage reporting

  • DD: Bill is filtering the boron out manually on his end, so he has a workaround for now

• Issue here:

• SMARTS-based filter can be used for this; likely will make this a new step between compute and analysis

QCEngine torsiondrive

David

• JH: would be ideal if we could add torsiondrives as a procedure via qcengine.compute_procudere; would allow for usage in many places by many people

  • DD: I can aim for this instead of shoving everything into openff-benchmark; agree

• JH: td_api recommended route; gets you everything that QCEngine supports

  • DD: yes, we’ll build it out of td_api

Release

David + Jeff

• Release 2021.03.17.0 cut