2025-06-10 FF Fitting Meeting

2025-06-10 FF Fitting Meeting

Participants

  • @Lily Wang

  • Bill Swope

  • @David Mobley

  • @Chapin Cavender

  • @Jennifer Clark

  • Christopher Bayly

  • @Matt Thompson

 

Discussion topics

Notes

Notes

LW: Parameters are exploding with loose priors

  • CB: Could it be that the gradient is very large, or is it that there are two mutually exclusive options being pursued simultaneously? Is there a systematic pathway to determine the cause?

  • LW: My initial thought is that the step size is too large. We should shoot for the loosest parameter possible.

  • CB: Your plan sounds good.

BS: OpenFE is starting an interest group on partial charge models if anyone is interested.

  • CB: Are they guessing and checking or drilling into the “why”

  • BS: They will likely be focusing on the cases where OpenFE does worse than FEP+

  • CB: Will they have virtual sites

  • BS: There are a couple models of interest but yes

  • CB: I came up with ELF to accommodate having different partial charges in different environments

  • BS: We are interested in if there is a common thread between the edge cases (e.g., charge transfer), this knowledge will inform on next steps.

  • CB: I think training NAGL on ESPs instead of AIM1BC because of its defects. Is there progress on that Lily?

  • LW: Talk to Danny Cole about that, they have NAGL-MBIS

  • BS: We’ve had one meeting and have a focus on binding FE instead of salvation.

  • CB: If they are doing a full binding FE, I expect that the salvation will show up. Nonetheless, salvation FE would be fast so might as well. Actually a PB calculation could be a good quick litmus test…. cool!

  • BS: I’d like to get a get of grand problems from these discussions

  • CB: Keep in mind that if a partial charge is altered by polarizing, there’s an energy cost to that that must be accounted for. I had a collaboration with MG about this (Willa’s work).

  • BS: Are virtual sites more important than direct polarization?

  • CB: Does OpenFF have a virtual site FF?

  • LW: We do, but OpenFE doesn’t support it yet

  • CB: Virtual sites on water is lower priority than on solutes (except Schrodinger says they aren’t important for proteins)

  • LW: It doesn’t seem like OpenFE has plans to incorporate polarizability

  • CB: CC would our plug in work with OpenFE?

  • CC: I don’t know enough about that, but Willa has done solvation free energy so that should be possible

  • CB: It seems that neither virtual sites nor polarizability are an option

  • LW: They create the system, so they would need to be able to support arbitrary forcefields for that to work

 

Action items

Decisions