2020-10-02 QCA Submission Meeting notes

Owned by David Dotson, created
Last updated: Nov 03, 2020 by Joshua Horton
3 min read

Date

Oct 2, 2020

Participants

  • @David Dotson

  • @Trevor Gokey

  • @Pavan Behara

  • Ben Pritchard

  • @Joshua Horton

Goals

  • New advancements

    • QCArchive now at QCFractal 0.14.0!

      • INCOMPLETE remediation path, performance improvements for inserting tasks

    • QCEngine release forthcoming

  • New submissions

    • Protein Fragments TorsionDrives

    • Enamine REAL subset Optimization

    • Jessica Maat’s dataset

    • PhAlkEthOH dataset

    • Protein Fragments Optimizations (resubmit with version bump, constraint indices 1-based)

    • XTB methods on JACS set

  • Upcoming infrastructure improvements

    • STANDARDS-based versioning #137

    • Identifier support in QCElemental.Molecule?

      • resolved our current solution last week

  • Upcoming science support

    • Selection of TorsionDrives from a dataset by SMARTS matching

    • PCM-based implicit solvent pathway

    • ESPs and wavefunction storage

      • will examine next week; there are known Psi4 blockers that we need resolution on

    • Uploading datasets calculated on private server

  • Larger advances

    • Automated FF coverage gap identification, torsion prioritization, submission generation

    • Benchmarking (dashboard, etc.)

Discussion topics

Item

Presenter

Notes

Item

Presenter

Notes

ANI harness problems

Trevor

  • DD: need a QCEngine release to propagate the fix, not out yet

QCArchive performance improvements

Ben

  • BP: previously 3 database commits per submission molecule; now 3 database commits per submission

  • TG: will this break submission that uses multiple threads?

  • BP: not sure, might have weird race conditions

  • DD: Josh, are changes required for QCSubmit’s submission to support this?

  • JH: just use one thread for QCSubmit, shouldn’t require changes

restarting INCOMPLETE tasks that with data

Ben

  • TG: I will restart all INCOMPLETEs that have data across all OpenFF datasets; getting list of INCOMPLETEs from Ben to intersect with

  • DD: One-off restart would be informative for us to see if this problem is fixed

  • BP: All of the ones from this year are optimizations

Conformer generation

David

  • DD: what’s the best approach to choosing a subset of conformers for torsiondrives from the optimization set?

    • Molecules have on the order of 30-40 conformers; probably don’t want to torsiondrive them all

  • JH: will add compressed dataset.json support to QCSubmit

  • We will proceed with all of them; after the initial optimization the TorsionDrive service will choose sufficiently unique starting points for wavefront propagation; shouldn’t be as large exactly as all the conformers

XTB on JACS

Josh

  • JH: Need xtb environments to support

  • DD: will add xtb to kitchen sink prod env, as well as give it its own targeted prod env; JH will get version desired to DD

Torsiondrive SMARTS matching

David

  • DD: both Jessica Maat and Dominc Rufa have this need; I have committed to making a PR against QCSubmit with a prototype of this functionality

    • JH: how would this work?

    • DD: don’t have a clear idea yet; but we do want a minimally complex set of tools that enable this for users like Jessica and Dominic

  • PB: where are SMARTS patterns stored on the TorsionDrives

    • TG: At the entry

PCM

Josh

  • JH: you have to restrict the symmetry to C1

    • going to update psi4 to automatically set the symmetry C1

  • Conclusion: if psi4 forces symmetry to C1 on PCM usage, then that allows us to submit specs that use PCM (distinct from default spec)

Protein Fragments

Josh+Trevor

  • TG: we don’t actually want to constrain dihedrals for torsiondrives

  • JH: QCSubmit ignores them anyway; we really don’t want these

  • TG: Will ask DM for clarity on whether we want both a constrained and unconstrained Protein Fragments optimization dataset

  • JH: Will prepare an constraint-indices-fixed version as a PR so we’re ready when we have word

Uploading datasets

Trevor

  • TG: not a priority at the moment; more a want than a need right now

Automated submission

David

  • DD: a bit far-reaching, but we’re interested in a decision loop around QCArchive for improving parameter coverage of our forcefields

  • TG: Need a data access layer first; will depend on this

  • DD: Agreed, we can revisit this later; something to think about later

Action items

@Trevor Gokey will restart all INCOMPLETES that have data across recent OpenFF datasets, getting list of INCOMPLETEs from Ben to intersect with
@David Dotson will proceed with submitting all conformers in Protein Fragments TorsionDrive dataset; leaving out constraints entirely
@David Dotson will add xtb to kitchen sink prod env, as well as give it its own targeted prod env
@David Dotson will create a minimally-complex set of tools to enable SMARTS matching on TorsionDrive datasets for Jessica, Dominic as a PR to QCSubmit
@Joshua Horton will pursue psi4 forcing of symmetry to C1 on PCM usage with Ben Pritchard
@Joshua Horton will prepare a constraint-indices-fixed version of the Protein Fragments Optimization dataset
@Trevor Gokey will consult with David Mobley on interest in an unconstrained Protein Fragments Optimization dataset

Decisions