2025-03-06 Protein FF meeting note

BPS connections

@Chapin Cavender

• Markus Miettinen / NMRLipids workshop

  • Want to start including data on membrane proteins. This would be congruent with our efforts on protiens, asked if we’re interested in participating (planning/discussion of needs, possibly attending workshop in Norway this summer, or helping with implementation).

  • MS – Maybe JH should go, could be a good representative since she’ll be taking over membrane protein stuff from AF.

  • CC – I’ll loop you in by email.

  •  

• Anand Ojha / Sonya Hanson / Flatiron benchmark of halogenated ligands

  • Getting bad agreement with expt from halogenated compound binding FEs. Wondering about supporting them using sage.

  • MS – Two sides to this - halogenated phys props (and therefore our training) might be off. Also might be sampling issues.

  • LW – Do you know what they’ve tried? Have they tested OpenFF yet?

  • CC – I don’t think they have. They tried with GAFF and CGenFF. Want to try with OPLS4.

  • LW – Maybe they’d be more interested in talking with OpenFE?

  • JW – I’d be in favor of telling them to try it out and offering an hour or two of support.

  • LW: info@openforcefield.org will get the lead team

  • CC – Flatiron has a lot of visibility in our field, I think they’d be a good way to get our name out in the field.

  • JW – They’ll probably want to end up talking to OpenFE, but let’s start contact using info@openff and make an introduction from there.

  •  

GB3 NMR fits

Open

@Chapin Cavender

• (slide 8) MS – Fact that uncertainties in third row don’t match observed inter-replica variability suggests that sampling is insufficient

  • CC – Agree

• JW – What does it mean that the candidate FFs look better or worse depending on the strength of the restraints, but ff14SB looks good with the original (tighter) restraints?

  • CC – (might just be that ff14SB is a good enough FF to not be affected by overly strong restraints, whereas our candidates aren’t good enoguh to overcome)

• MS – Complicated question (see recording ~22 mins)

• CC – I’m close to getting replica exchange working on UCSD cluster, that should help with sampling.

• MS – I’m still curious about seeing predicted vs. actual FE for the green-to-purple generation

• JC – Key takeaway might be that multiple confs might have same %native contacts, and they weren’t able to connect to each other?

  • CC – Right, things were stuck in wells in native contact space, but we weren’t able to sample between them. But with looser constants you can move between barriers and connect those. And repex will help even more.

• LW – IIUC, with better sampling, the green curves look like the AMBER curves, but GB3 still falls apart.

  • CC – Right, and if we look at chi2 for unbiased sims, the sampling for the green curve looks better than for our other candidates.

  • LW – Just kinda surprised that the scale of the well seems similar but we still see unfolding.

• JC – You fit purple using green… does it worry you that the purple slope is so much shallower than green?

  • CC – Yes, to some extent, and I see the chi2 value being higher (which is bad) as well. But worth keeping in mind that ff14sb likely OVERstabilizes the native state, so the best chi2 might be attained with a shallower curve than ff14sb

  • MS – … (mostly agree)

  • CC – And I expect repex to improve this even more.

• MS – Also on slide 3, we see some depopulation of some windows since the restraints aren’t strong enough to keep them in.

  • CC – Yes, …

• MS – What are downsides to just reweighting to GB3 NMR?

  • CC – We risk overfitting to protein, and errors in NMR data will be baken into parmaeter set. I saw some other work at BPS where peple had trouble with GB3 and not other proteins.

  • MS – Worth doing umbrella sims for other proteins?

  • CC – That could be a good idea. Straightforward to set up.

•  

• AF – On new QM dataset - I found an issue in how the PDBs were sequenced (also wrt modified residues, especially selenomethionine).Dataset is almost ready, doing some cleanup suggested by Jen then it should be good to go.

  •