2020-12-07 Core Developers Meeting notes

Roundtable updates

• DH – Worked on openff-benchmark infrastructure. Implemented analysis tools, working on plotting. Also integration into CLI. Babysitting calculations to get free energy jobs done by the end of the year. Writing LiveCOMS best practices article for benchmarking.

• SB – Wrapping up recharge <--> QCArchive interface. Now can generate lots of ESP/electric field data. Tested locally and it looks good. Updated nonbonded package to work directly with QCA datasets. Doing automated optimizations of ES properties is now really easy. Used this infrastructure to redo BCC study, so now have BCC data refit against some QCA dataset, and XtalPi has tested it. Made espaloma-like project called NAGL that can reproduce AM1 WBOs. Also analyzing data from XtalPi to see how it agrees with our nonbonded refitting, things seem to be moving in the right direction. Looking at how to generate training sets from molecule fragments, seems very powerful.

  • NAGL – – Uses same method as espaloma, and can be trained on the scale of minutes (after partial charge calc). NAGL doesn’t use gimlet/espaloma networks, but rather handles training and usage of new ones.

• MT – Cleared and pushed on some toolkit PRs. Updated to MacOS 10.15+, which was a mistake. Had espaloma interface meeting. Discussions on units and future common unit infrastructure, testing some different designs. Worked on System potentialhandlers, particularly tricky to implement torsions. Some friction on handling units while spec isn’t determined, but in the short run I’ll do everything in implicit units. Worked on polishing toolkit showcase example – instructions on installing extra deps can be added to the notebook.

• DD – A bit scattered last week. Met with AWS spot team to talk about using it for QCArchive. Will have a followup meeting soon. Goal is to be granted access to unused capacity on AWS for our QCA calculations. We’ll have to decide which institution is the account holder – DM proposes UCI. This week I’ll review PB’s QCA dataset submissions (PB says there’s no hurry). Will be focused on finishing unit tests for openff-benchmark. Will be pushing for merging PRs in openff-benchmark. Will also submit partner datasets into QCA.

• JW – Debugging the following code snippet.

  • from openforcefield.topology import Molecule mol = Molecule.from_smiles('C=C') mol.generate_conformers() mol.to_file('xxx.sdf', file_format='sdf') mol2 = Molecule.from_file('xxx.sdf')

  • Currently benchmarking validation has about a 5% error rate on RDKit backend.

  • Also need to do more for conf generation – apply RMS cutoff to user-provided mols.

• PB – Working analysis of WBO interpolated parameters, and seeing how quality of fit changes. Started doing FF comparisons using some new analysis scripts. Also worked on genentech dataset submission, looking at issues with fragmented molecules, possibly due to some duplicate 3D structures.

  • Some torsion energy plots (not final, bugs possible):

0.8.1 release