2021-04-14 Benchmarking for Industry Partners - Development Meeting notes

Torsiondrive execution testing

Lorenzo

• Seeing issues on Janssen cluster; appear random, consistent with network-filesystem latency failure modes

• Will try the approach taken for optimization benchmark; install conda env to each compute node local filesystem

Updates from team

• DH: ffbuilder runs even if no params directory already present now

  • LD: could 2021 version work with this?

  • DH: could work, might be issues; if folks use different versions of schrodinger, get different ffs in the background

    • 2021 just used OPLS4, nothing else really different

    • might consider changing the tag to include the OPLS version

    • DD: is there an easy way to get the OPLS version out of the tools? Can then have opls-<version> and opls-<version>-custom?

  • DH: short interactions with Bill, worked with Lorenzo on a tour of openff-benchmark repo, next steps with analysis

• LD: starting to deal with analysis that Xavier, Bill requested

  • got a tour of the existing analysis from DH, looking to make use of what we have to accommodate Xavier, Bill needs

    • going forward with building both analysis, then will decide whether to include both, or to choose one

  • trying out torsiondrive executor, put testing results on Confluence

    • testing classes of molecules known to be difficult for torsiondrives

    • https://openforcefield.atlassian.net/wiki/spaces/~954728248/pages/1518010447

    • N(amide)-C(arom) rotatable bonds set, known to be difficult case for many FFs

    • JW: interested in doing a working session at the end of this meeting if we can for identifying which torsion to drive

    • DH: might need to consider kekulization too?

  • JW: rotatable bond definition in RDKit:

  • Case in point on rotatable bonds being kinda dodgy:

    • from openff.toolkit.topology import Molecule mol = Molecule.from_smiles('[O-]C=CC=CC=CC=CC=O', allow_undefined_stereo=True) print(len(mol.find_rotatable_bonds()))

  • DD: will add driving of every rotatable bond to torsiondrive executor for further investigation of limitations

• JW: reviewed and merged Josh’s PRs; cut the release on Friday 4/9

• DD: worked with Jeff to cut the release, instructions out to partners for export and submission

  • DH: perhaps folks are waiting for Schrodinger pathway?

  • DD: will ping partners on Friday for updates

Schrodinger pathway

David H

• DH: fix is in for missing opls-dir in ffbuilder command; also simplified input for ffmerge

  • See a way forward on attaching OPLS version to outputs so we can support multiple versions in principle