Page Comparison

"Include a small subset (~100 molecules) of small molecules relevant to OpenFF (for distinguishing levels of theory etc below)"

Has this been done? [Ask Chodera Lab]Need variety of metal centers and organic moieties to achieve this better to have large swath at low level of theory

Decide on Level of Theory

...

•  Is BP86 / def2-TZVP for primary and B3LYP-d3bj for final good enough?

Search through existing databases

Use Gemmi to handle dataset

•  PDB Chemical Component Dictionary (CCD)
  [Brent achived]
•  download and filter (Brent)
•  Submit dataset
  figure to out what's going on with our PDB dataset, all issues are SCF convergence
•  tmQM [paper, dataset] (which sources from CSD) - 86K transition metal complexes
  Utilize this existing dataset to begin testing out NNP strategy regarding encoding spin state all spin states = 0.
  Currently being worked with by Chodera lab, should get into QCArchive at some level
•  Crystallography Open Database (COD) – CC0 licensed
•  CSD (cambridge strucural database) filtered for stable small molecules, size, elements
  might need to discuss release of subset as open data
  [Repeat of tmQM?] Look for structures neglected by tmQM of interest?
•  MPtrj: Materials Project Trajectory Dataset

...

•  Any transition metal for any transition metal, let it run and see what happens (will get something relevant most of the time).
  Switching within column is almost always okay.
  [Will need in house tmQM for this]
•  substitutions: H->F, Ph->pMeOPh.
  Consider RDKit has "replacesubstructs" method
•  Use MD to search for transition states that are hard to find with conventional DFT, then minimize for a few steps. Better than rough conformations from RDKit. [Pursued by Chodera Lab]

Conformal search:

•  take from CIF files (CIF is the new PDB)

CMILES Issue: Organometallics are difficult, not supported

https://chemrxiv.org/engage/chemrxiv/article-details/676050a56dde43c9085b4ccd

•  Jeff thinks cif -> rdkit -> qca
  - not sure that's necessary because QCArchive has a hash to represent the molecules.
  - Don't have to use CMILES as the name (i.e., index), can be arbitrary.
  - QCArchive doesn't need CMILES, Prepare to pair program with JClark on cif --> QCA pipeline by bypassing openff molecule structures that are dependent on CMILES and directly compare cif files to QCArchive
•  RDKit had an organometallic class when assessing implicit hydrogens, reverse engineer an expression?
•  From 09-05 notes, CI: this is something we’ll need to consider for QCArchive too. It’s one Record per conformer, so we need to be able to associate a metadata record with a record. [
  - QCArchive has a metadata file field to add such information]. Also openff Molecule.conformer is an attribute containing pint.Quality class with coordinates and units. However, the list seems to usually only contain the coordinates of the molecule of main molecule.

Computed/stored properties

...

•  Make a new dataset type for these properties, run opt as this dataset
•  Determine the keywords for psi4 to obtain these, change provided output? (Note tmQM used Gaussian NBO analysis for these, where output is trivial.)
  OptimizationResultCollection.create_basic_dataset pull final geom from opt and then create input for qca SP with these properties.