2025-03-19 Cole Lab Check-In (Aus) Meeting Notes

LJ Fit for NAGL-MBIS Models

Lily Wang

• LW: Fits converged in the sense that new iterations very slow, but objective may not be converged

• LW: Options for benchmarking - Sage benchmark set (needs old Python env) or new OpenFE benchmark set (but this may be slow)

• DC: Could use absolv as done for double exponential paper

• DC: May be worth checking overall errors on the training set as a rough guide to current performance

  • LW: Will post plots to chat

• LW: Could do additional valence fit in ~ 1 week or so

  • DC: Keen to take this into the SMEE-fitting work

• DC: Can we transfer Overleaf to OpenFF/ somewhere others can make edits (David Mobley to be involved as well)?

  • LW: Unsure if OpenFF has subscription

Summary of Cresset Issues and Plans for Bespokefit-SMEE and SMEE Valence Training

Finlay Clark (see slides)

• LW: Difference between SMEE and Sage on the tricyclic molecule: We don’t fit to in-ring torsions with Sage - gives high force constants.

• LW: On plans for SMEE valence fitting: Would be cool to see results with fitting to new tripeptides/ dipeptides dataset