List meeting participants using their @mention names:
Item | Presenter | Notes |
|---|---|---|
preliminary fitting result update | HJ | |
benchmarking relative to QM discussion (general) | DLM asking | Frank Pickard interested in discussing QM level of theory/perhaps assisting. |
1. HJ
CIB: better to check points with decreased WRMSE (case where new ff better describes than the existing ffs)
DM: good news that the new approach of improving parameter (by using better training set)
CIB: contribution of QM vibrational frequencies in fitting? How to quantify the contribution? (lower priority, maybe after May meeting)
can check how strongly vibrational frequencies attribute during the fitting
fitting w/ and w/o vibrational frequency targets and compare the performance?
2. Theory benchmarking
DM – HJ+LPW, Frank Picard at Pfizer is an experienced QM guy, and wants to follow our studies about levels of theory. What’s happening there?
LPW – We haven’t made a ton of progress on this
HJ – I’m working on an initial set
DM – You should tag Frank Picard and let him know what you’re up to
DM – LPW, did you and DC talk about standardizing QM methods?
LPW – B3lyp B3Bj(?) dispersion correction. He watns to tuse a bigger basis set than we use. We wanted to look into using this for single point energies/gradients.
LPW – He’s interested in more complex FF forms, tabulated potentials, other ideas
DM – We cna look into that in the longer run, but need to focus on importing amber biopolymer FFs now. He can figure out which fragment to use, but if there’s he’s interested in different basis sets, let’s include them in our benchmarking
LPW – DC suggested a clever way to penalize intramolecular hydrogen bonds. This is by adding a penalty function which is high from 0 to 3 angstroms (between hbonding groups), and 0 after that. We may want to follow up on this
CB – Even if parts of our training set are excluded now due to quality issues (like hbonds), we should tag them appropriately and “titrate” them in for subsequent force field fits, to make our training set capture more difficult cases. ELF-style conformer selection may help here.
JW – Checking in on timeline for fitting:
JW – Aiming to have QM data ready April 20-23
Valence people – Estimate for runtime? – 4 days
Phys prop people – Estimate for their runtime?
Do we want to do a nonbonded refit? For some clearly improvable set of atoms?
Probably not
We won’t plan on nonbonded refitting to happen for the 1.2.0 release
Add action items to close the loop on open questions or discussion topics:
Type /decision to record the decisions you make in this meeting: