2024-03-06 Meeting notes

Moving project to org on GitHub?

MT

(General): Should do soon

MT: Will coordinate with JW

LW: Will this be user-facing?

MT: Yes, not sure about the details. Should have some minimal docs. Re-visit in a couple of weeks.

Trello update

MT

• https://trello.com/b/dzvFZnv4/infrastructure?filter=label:benchmarking

DDE definition?

• LM: PB/SB’s code has a different way of computing DDEs, might be preferred. (Current code compared MM energies to the conformer with the lowest MM energy). Some issues:

  • Lowest-energy MM might not be the lowest-energy QM conformer

  • Always defines ddE as lowest-energy conformers as 0, which dilutes the 0 bucket in the KDE/histogram

• A better way would be to use the conformer that has the lowest QM energy as the “zero” point even though it is not guaranteed to have the lowest MM energy. This should be feasible since it’s a matter of looking at a different conformer index

• Q: Would it be useful to have this as a toggle?

  • Using the QM minima is probably better, but might be useful to have both

MT: How much does this impact results?

LM: Different methods are qualitatively similar (“is this force field better?”) but quantitatively different compared to the 2.0/2.1 benchmarks that already exist.

BW: Should we just exclude all self-self dde? They’ll always be zero?

LW: Yes

(Currently these are all zeros - maybe they should be NaN/nullables?)

LW: Any case in which a NaN is the actual result of a dde calculation (as distinct from setting the minimum to NaN)?

• MT: Probably not? But maybe

MT: Could have a better way of filtering it out? An extra attribute in the DDE that “tags” the minimum as being a minimum, and not some very small dde approaching machine precision/rounding