Page Comparison

...

11 Dec 2020

Participants

• David Dotson

• Trevor Gokey

• Pavan Behara

• Ben Pritchard

• Joshua Horton

Goals

• New advancements

  • STANDARDS-based versioning #137

• New submissions

  • PB: Genentech OptimizationDataset (ready for review)

  • PA/JH: Protonation/tautomer state enumeration dataset (ready for review/ needs name update)

  • Hessians/wfn on CHO sandbox set

  • Dataset consolidation for Hyesu benchmark sets

• Upcoming infrastructure improvements

  STANDARDS-based versioning #137

  • Dataset index on qca-dataset submission #147

  • Do we need to change the index system we use for molecule submissions?

    Github link macro
    link https://github.com/openforcefield/qcsubmit/issues/61

  • Github link macro
    link https://github.com/MolSSI/QCFractal/issues/650

  • More targeted error cycling; what else do we need in the report for decision-making? Github link macrolinkhttps://github.com/openforcefield/qca-dataset-submission/pull/165Stale jobs fix on QCFractal server/managers

• Upcoming science support

  • Enforced c1 symmetry in psi4 is almost ready

    Github link macro
    link https://github.com/psi4/psi4/pull/2037

• Larger advances

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

  • Benchmarking (dashboard, etc.)

Discussion topics

...

Item	Presenter	Notes
Stale job updates	Ben	Ben is planning a update that should fix this in future e.g. manager.manager_name: PacificResearchPlatformMM inconsistency with how this is stored server-side (cluster)
Database error provenance	Ben	Error records grow monotonically in database; any reason to keep them? should we delete ones that have no references to them elsewhere in the DB? TG: perhaps the sequence of errors should be preserved if we have yet to complete a result? only delete the sequence of errors once it’s complete?
STANDARDSv3	Trevor	DD: Trevor presents STANDARDS, discuss comments addressed, open floor discussion before we move to adopt TG: CMILES provenance info still unresolved; looking for a solution that doesn’t introduce complexity on top of QCArchive’s built-in approaches if possible DD: Not sure it’s possible to achieve the information structure we want without introducing some complexity on top, handled via a data access layer that encodes these STANDARDS BP: Thinking on an approach to support this more deeply within QCArchive Thinking of cases that require changes QCSchema, will require some thought around the whole stack DD: I don’t think this is a blocker for STANDARDSv3; think we could establish a working group for improving graph molecule support in the whole QC* stack, which would translate to better MM and ML support more generally BP: there is an MMSchema effort in MolSSI; this may be a good route to pursue for building out a representation that can be consumed by QC* JH: Tried building some WBO estimators from wavefunction data; not straightforward, relies on finnicky cutoff choices BP: thinking of building a table that stores connectivity, where there can be multiple records for a given molecule, storing provenance info for the tool that generated/interpreted out that connectivity from the geometry, wavefunction, etc. DD: think there is an opportunity to build a library that ties together QC mols to MM mols more generally, given interpretations from tools like RDKit, OpenEye, etc. Could then be used to populate the table that Ben suggested TG: connectivity, formal charge, if we had those, really could make the whole graph; could then use that instead of CMILES if desired 3 ayes, 2 abstain Adopted! JH: happy to walk through doc, bust out into issues primarily on QCSubmit, put together into STANDARDSv3 project board for execution
		DD: will schedule working session with Josh, Hyesu for consolidating benchmark datasets, adding mols

Action items

•  Ben Pritchard will pursue resolving inconsistency in manager.manager_name in Manager configuration vs. cluster key stored server-side
•  Ben Pritchard will consider an approach to pruning old, unreferenced errors accumulating in server DB
•  Joshua Hortonwill walk through STANDARDSv3, translate components that need to be implemented yet into issues on QCSubmit, qca-dataset-submission automation, etc.; will set up a project board on QCSubmit to parallelize implementation work with others
•  David Dotson will schedule a working session with Hyesu, Josh for consolidating benchmark datasets, adding molecules

Decisions