2020-03-26 Software Scientist Onboarding notes (Thompson version)

Open Force Field onboarding day 3

Wagner
Dotson

Health of the infrastructure

• QCSubmit

  • Doesn't exist yet, being built by Horton, who is very careful

• QCPortal

  • Best corner of the infra, beautifully built by Smith @ MolSSI

• QCFractal

  • In decent shape

  • Many issues with file systems

• Property Estimator

• ForceBalance

  • Frequent pain point

  • Many heterogenous interfaces, file I/O, etc.

  • Fundamental issue of representing molecules: can't have all 3

    • QM - nuclei, charges, not really much else

    • cheminformatics - "graph"-like object, atoms, elements, bonds,
      sterochemistry, aromaticity (everything needed to type using SMIRKS)

    • MM - parametrized system (i.e. including force field)

• CMILES

  • GitHub - openforcefield/cmiles: Generate canonical molecule identifiers for quantum chemistry database

  • Bridge between QM and cheminformatics (for the interop between QCA and the cheminformatics that ForceBalance needs)

  • informatics has "allowed" operations, i.e. SMIRKS searching, serialization

  • this bridges some operations into QM world

  • QCA has particular criteria for defining two molecules being the "same", which may differ from informatics definitions

    • QCA criteria is coordinates being similar enough (for purposes of looking up in the DB if a molecule already has QM data)

    • http://docs.qcarchive.molssi.org/projects/QCElemental/en/stable/model_molecule.html#molecular-hash

  • Bridge breaks some small % of the time

    • i.e. proton transfer in QM optimization, resulting in a different informatics

      • would also break MM parametrization

    • stereochemistry

      • charge method in MM world can depend on sterochemical conformation

      • "stereogenic groups"

      • can be covered in SMILES, but can change during QM, breaking the bridge

        • i.e. scanning over a trivalent nitrogen, flipping its chirality though angle of planes

          • Useful barrier to study in MM world, QCA can do the torsion drive, CMILES representation is different before and after

          • Further edge case of being completely planar, i.e. between chirality

Lots of scattered glue for various conversion, some value in unifying at least conversions? I.e. any QM->MM representation is done in the same way across people, projects, labs, etc.

Dotson raised an issue on the QM->MM conversion weeks ago in which Wagner detailed the problem, had a hard time finding it

Build process

• gh/omnia-md/omnia-recipes

• Many deprecated recipes, not really a problem since most things don't depend on each other

• Goal is to migrate everything to conda-forge, no small feat but Jaime is making great progress

• Major pain point is OpenMM, legacy reasons based around OpenMM's build system conflicting with conda-forge's preference

• Once OpenMM can migrate, the levee breaks and the rest of the stack can migrate (hopefully!)

• openforcefield #286: Wagner wrote thorough docs on the standard release process

• Another time: we will work on improving developer docs (perhaps on another wiki)