Skip to end of metadata
Go to start of metadata

You are viewing an old version of this page. View the current version.

Compare with Current View Page History

« Previous Version 5 Current »

Date

Participants

Goals

  • Updates from MolSSI

  • Report on current queue status, and whether managers need updating

  • Open forum for user questions

  • Wavefunction investigation update (Hyesu)

  • Protein fitting submissions (Cerutti)

  • Additional datasets?

Discussion topics

Item

Presenter

Notes

QCElemental

Hyesu

  • When we try to return the wavefunction from QCEngine, there are protocols we can use to set the output.

    • only orbitals and eigenvalues work

    • other protocols fail to return the result

  • Only ”orbitals_and_eigenvalues” appears to work

  • Josh: this might actually be the only option that works

  • David: should be good to proceed then with ”orbitals_and_eigenvalues”. Hyesu should be clear to proceed with that and see if it meets our needs for calculating ESPs downstream

Initial Protein Dataset v2

Josh + Dave Cerutti

  • Josh: do the number of unique molecules, number of conformations match with your expectations?

  • DC: sounds like you have the whole thing; good to go!

Disaccharides

Dave Cerutti

  • Added conversion from Angstrom to Bohr

  • Generating initial 1000 molecules, tar+gz-ing them up and pushing them to PR

ANI

Josh

  • We want to add ANI to the ligand dataset as soon as possible

  • Need to make some small modifications to QCSubmit to handle not having a basis submitted in an ANI spec

openmmforcefields

Trevor

  • openmmforcefields drops all the hydrogen constraints for MM calcs

    • this happens to be what we want right now

    • in the STANDARDS.md, we should specify that we use the _unconstrained version.

Prod environments

Trevor

  • we should version pin the ANI components where possible

Expanding submissions

David

  • Agree that we can probably view additional compute beyond the original submission as a bonus, decoupled from the lifecycle of the original dataset (because it will be decoupled in our lifecycle model).

    • This is a good thing, because more experimental compute regimes will likely show new failure modes, and may never be clearable; this shouldn’t stop us from considering a dataset “complete” for archival/provenance purposes.

    • Allows us to experiment without worrying about holding up a dataset’s more rigorous advancement with established compute specs

Action items

  • Hyesu Jang will continue to develop our pathway to ESPs from wavefunction results with QCEngine
  • Joshua Horton will make necessary modifications to QCSubmit to handle ANI specs (e.g. no basis)
  • Trevor Gokey will add a clause to STANDARDS.md specifying our use of _unconstrained OpenFF forcefield versions in future specs
  • David Dotson will pin versions of torchani, pytorch in ANI deployment
  • David Dotson will confirm agreement on approach for expanding submissions; propose a pull request into STANDARDS.md, LIFECYCLE.md.

Decisions

  • No labels