2021-08-02 Core Developers meeting notes

CC

• QCA datasets

  • Met with Dotson last week on submitting datasets to QCArchive. Ran into a problem where psi4 removed the version that we were using from http://anaconda.org , so we will need to update. So this is a speedbump to getting set up on the Triton cluster at UCSD. There’s also a conversation we should have about how to get QCA workers set up on the triton cluster using available conda packages, in a reproducible/consistent way with other QCA workers.

  • JW – I could probably rebuild the lost psi4 packages using their conda packaging machinery – it’d be better to avoid needing to do this, but we could take this route in a pinch.

  • CC – DD is going to try a few things to get around this, and work on a policy for how to update the environments in a sensible way.

• CC – What’s our preferred format for storing multiple conformations?

  • JW – SDF is best for us. We never assume that any input is multi-conformer, since few (if any) file formats have a spec to indicate whether multiple entries are multi conformer vs. multi-molecule

• JW – I’m behind on biopolymer infrastructure. CC, is this blocking your work?

  • CC – Right now I’m focusing on QC datasets. I’ll let you know if missing infrastructure starts slowing things down.

• CC – Found a useful function in OpenEye where they have rotamer libraries already included - Can use this for dataset generation. However it doesn’t catch N-terminal glycine. So I may need to manually add a way to match these and tag the torsions myself.

  • JW – I’m not sure, this may be hard to fix because everything’s black-box code.

  • CC – There may be an open-source option in mdtraj, and the rotamer libraries are on GitHub.

• CC – Should I contact OpenEye?

  • JW – Sure, I wouldn’t spend too long on it though.

MT

• Took a break from vsites, worked on interpolated parameters. I think this is all done (handling torsion k, bond k, and bond length). I’ve made unit tests and these are looking good, energy tests are at an acceptable level (though this isn’t a high bar, since sqm gives different results wiberg bond orders on different machines). Right now error is about 0.001 kJ/mol. Would like to get internal users to give it a spin.

• Started implementing SMARTS matching improvements suggested by Lily in OFFTK. This is priority #1 for me - Most of my CI time is spent waiting on AM1 results because the librarycharges are so slow.

• Worked with Josh Mitchell on coordinating docs changes. Moving forward with the design proposal that he put in slack re: “Book (user guide) vs. Reference (autogenerated API reference)”. I’ve gotten to the point where everything HAS a docstring, but I’ll be adding more and starting to work on the Book now.

  • I could use input on what docs people want. So I’d like to start getting feedback on this.