2020-08-03 Benchmarking meeting notes

Background

• Generally, we want a way to reproduce the plots in this blog post and repo

  • https://openforcefield.org/news/benchmark_small_molecules/

  • once there’s a new ff release, “hit a button” and re-generate data

Can we use QC* for the compute

Need to make sure the MM calculation is sampling the same space as QM torsion drives
QCA is designed for rigidly-specified calculations, will not be very reactive to frequent modifications of the shape of the analysis.

• Can a modified form of QCEngine be more adaptable to more diverse operations?

  • Yes, but would require a fair amount of work (bunch of harnesses)

Trevor has some infrastructure for this

• Fetches QC data from QCArchive, parametrizes structure with MM and tags MM and QM structures

• “static” calculations are doable, exploratory calculations are not well-suited for QCArchive

  • basics (energies, RMSDs) feasible within existing infrastructure

    • minimizing MM geometries introduces compute limitations and some stochasticity

  • exploratory (“what happens if I modify the ddE formula, or do this other thing?”) not really questions we’d want to shove through QCA

What properties should be included?

• Simple properties doable short-term

  • ddE

  • RMSD

• More complex analyses probably not tractable for automation