We will start a confluence page to spec out a study which will explore co-optimizing both the VdW parameters and ‘charges’ (BCCs, templates, charges directly) against a limitied training set of compounds / properties.
Agree that using direct Pol is okay but should maintain ability to run, and thus compare with, SCF. Should also compare to Bernie Brookes Method
Consider fitting to QM electric field vectors, not just potentials
Need to generate models which work across all pharma molecules from the start, by including S, Br, I, Cl, etc.
Initially, we can do BCC opts agains QM gas phase/implicit solvent fields. Later can tune to condensed phase (with restraint to QM BCCs) as currently being done for fixed charge BCCs