2020-09-14 QCFractal User Group Meeting notes

Updates from MolSSI

• Nothing to report

• Major QCFractal release ETA end of the month (~2 weeks, maybe longer w/ testing)

  • Major features: Getting jobs unstuck, metadata modification on jobs

• Looking @ 1 month cadence for releases

Queue/Manager status

• Last 1.5 weeks: Queue empty and compute turned off

• DD just error-cycled

• Working on/error cycling benchmarking submissions and ANI2x

  • ANI2x is flagship ANI job. about 120 complete of 480. Encountered infrastructure issues and are working through them

• Queue is open for more submissions

• TG is working on enamine submission

• DD is working on torsiondrives on protein fragment submission.

  • Will use QCSubmit

  • Will submit torsiondrives for

  • DC will join and learn to use QCSubmit – DC and DD will coordinate.

• Special thanks to TG for setting up costom environment for ANI jobs

User questions

• JM – Working with Chaya – Looking at getting MM TorsionDrive results in QCA.

  • DD – We can do that. Which FF?

  • JM – OpenFF w/ custom force field file.

  • DD – We may have trouble using local FFs

  • JH – Can’t do custom parameters yet, but can do released FF.

  • JM – How would I go about getting MM results using a released FF?

  • DD – Example here:

  • DD – Above link added MM energy evaluations to existing QM job

  • JM – Biphenyl torsiondrive dataset, with openff-1.2.0 and/or 1.2.1

  • (General) – Could do many MM forcefields

• TG – How would we do a custom MM FF in a local run?

  • JH – QCEngine OpenMMHarness would need to load FFs using the OpenFF toolkit. Loading path may go through OpenMMForceFields.

  • (General) – Inject full text of custom FF?

    • Con: Would be bad for provenance in public QCA. Since the same FF could be written as a sring or specified by a controlled file name

    • Pro: Would also be good for experimentation in local environments.

    • Solution: Could allow injection of custom FFs, but submission machinery could forbid use of custom FFs.

• DD – JM, let’s

  • 1) start work on submitting the new jobs

  • 2) start speccing out the use of custom FFs in local QCA jobs

    • BP – Could treat these as contributed datasets

    • JW – Might be best to steer towards, in a year, having datasets involving custom FFs be locally computable, and then upload the final results as a contributed dataset

    • JW – Would be good to keep track of process, especially how to preserve database for submission as a contributed dataset.

• TG – Is there a difference between a contributed dataset and a deepcopy of, eg, a torsiondrive dataset?

  • BP – Ideally, we’ll have an import/export feature set. But this will still suffer from having multiple ways to specify the same calculation.

  • JW – Is a contributed dataset indexed similarly to a normal dataset, or is it more like a tarball?

  • BP – It varies by dataset. They can be contributed at different “levels”

  • TG – Are they the same as a dataset object, or fully new classes?

  • BP – More of a flat dataset object.

  • (general) – Ambiguity about how to represent contributed TorsionDrive datasets.

• SB – Can we do implicit solvent using QCF? If not, what steps can we accomplish to make it possible? Do we have agreed settings for OpenFF implicit solvent?

  • JH – The psi4 nightly builds should be able to do PCM calculations.

  • (General) – Progress-wise, we need to be able to pass keywords to QCEngine, and ensure they pass through to psi4. That pathway should be implemented now

• SB – Progress on storing wavefunctions?

  • JH – This is blocked by a typo in psi4. We will be able to do this eventually. I’ve put this on Lori Burns' roadmap, and she should be able to resolve this on the ~month timescale.

  • TG – Is PCM just a modification to the keywordset option?

  • JH – Yes, through pcm_input keyword

  • JW – Have we settled on PCM? Is this due to availability or performance?

  • SB – We’re limited to PCM (two flavors: CPCM(?) and IEPCM(?)), folks haven’t really benchmarked to find the “best” option. Some knobs to turn like radius and cavity settings, also unexplored.

  • TG – Also, dielectric is another knob.

  • SB – Some pre-set “solvents” that you can choose, which set dielectric and other things

Update on /urgency of fragment mapping?

• JH – Currently have a workaround

• (General) – Bespoke workflow/fragmentation process works, but it’s closed source (requires OE) and requires the 2019 OETK.

  • !!! OETK WBO calculations always hang when called from fragmenter!!!

  • Fragmenter refactor may not happen until 2021 given current biopolymer/interoperability deliverables.