2023-10-20 Meeting notes

Datasets

• SPICE 2.0 at OpenFF level of theory

  • Peter Eastman’s stuff

  • other useful datasets to OpenFF et al.

    • e.g. PDB Chemical Component Dictionary – includes coverage of compounds like nucleic acids, but not in a targeted way

    • Enamine REALSpace : Enamine building blocks + representative robust coupling chemistries

  • philosophy is a lot is generated at OpenFF level of theory

  • PE hoping to combine 2.0 with 1.0 dataset

• nucleic acids dataset

• virtual sites fitting (e.g. sulfur, sigma holes, nitrogens, )

• electrostatic potentials and electric fields (possibly at a different level of theory?); Danny Cole / Josh Horton are looking into this as well

• general chemical diversity dataset (from ChEMBL)

  • OpenFF is currently fragmenting and clustering based on fingerprint similarity; quickest and easiest solution at the time

  •  

• Question: What kind of dataset is best to generate? OptimizationDatasets? TorsionDriveDatasets? MD-generated snapshots? MD-generated snapshots with a few steps of minimization?

  • OpenFF currently doing experiments to determine optimal datasets, with results expected in next couple of weeks

• OpenFF is currently lacking expertise in areas that pharma is interested in

  • Currently just looking at ChEMBL

  • JC – everyone is interested in enamine realspace

    • We have the SMARTS and SMIRKS strings for those

    • Patent spaces

    • We have scripts for scraping patent spaces

    • JC will share these