2020-10-30 QCA Submission Meeting notes

OpenFF BCC Refit

Simon

• Simon will add a directory of SDFs written out from the OpenFF Molecules used for the dataset, tar+gz it

• TG: should we include a default spec?

• SB: this is point calculations only

• TG: different workflow than we’ve been doing; is there a preference to use the omega stuff?

• SB: reason we use omega, we want openeye’s ELF10 implementation; rdkit doesn’t have this

  • would have to roll your own if you want to use rdkit for this

Protein peptide fragments constrained

Josh+Trevor

• Trevor: would like to be able to rename datasets; is that possible?

  • BP: if you pull down the dataset and resubmit, might work

  • TG: “OpenFF Protein Peptide Fragments constrained” v1.0 is good to go name-wise

• TG: Do we know what the rationale for the dihedrals DC wanted constrained?

  • JH: Major torsions in the three residues

(ANI, ANI1Cxx), ANI2x submissions

Josh

• JH: Are we even sure this is going to work, even with 3000 iterations?

  • Still seeing some that don’t finish running locally

• TG: will put together ANI2x supplemental compute

• JH: will do ANI, ANI1cxx

Enamine REAL subset

Trevor

• TG: not a submission yet, more of a toolkit-testing side project at this time

Unconstrained protein fragments

David

• TG: rename to “OpenFF Protein Peptide Fragments unconstrained v1.0”

• DD: will rename today

PEPCONF OptimizationDataSet

David

• TG: should follow the PhAlkEthOh pipeline, so can use that for dataset.json preparation

STANDARDS-based versioning

Trevor

• TG: working on this, want to separate mentally what it means to have a dataset standard vs. a FF-fitting standard

  • still have two weeks, feeling on track

INDEX

Trevor

• TG: I’ll review

• JH: will add the script used to generate the table

• DD: can add a GH action in a future PR to update at any frequency, commit to master

Ambertools pinning

David

• DD: does this fix our problems on Linux?

• TG: better to pin it than not to pin it with openmm+psi4

ESPs

Trevor

• TG: Use of AO density?

  • just sent a link of the wavefunction properties, have a script from Daniel Smith, but wasn’t sure if it generalizes to our problem

• JH: as of the last week, psi4 can save different parts of the wavefunction, “everything” option now works

  • will need to update psi4 in all managers to support this

• JH: will check for the “everything” wavefunction support in latest OSX conda package for psi4

• TG: Submitting the dataset from SB will give us a way to try out ESP reconstitution from stored wavefunction components

QCFractal advances

Ben

• Cleaning up databases, removing columns that aren’t used, etc.

  • still planning release next week; downtime up to 6 hours

  • everyone will need to update their managers

• Survey feedback: QCF is slow

  • considering putting a profiler in place on the server

• Should be a fix for multiple managers pulling down the same task

  • no easy way to test, but we’ll see if problem disappears or decreases in frequency

• Adding STDOUT from TorsionDrive service to the TorsionDrive record itself

• TG: still getting stale bombs

  • BP: initial thought for solution doesn’t help; could increase limit again; perhaps improve manager logic so it separates out bad results submission and tries to submit on its own

• BP: size of the body is really the problem

  • gets managers stuck; breaking up the results may be the answer, but would need to be engineered by us if not handleable by the choice of web server

  • BP: some are over 200MB

    • current limit is 175MB

• BP: I’ll increase the limit; may start seeing timeouts more often though

PCM

Josh

• Almost have PCM support in QCSubmit; need to add validations yet

• For Simon’s dataset, we’ll probably want to add a supplemental compute that uses PCM