2024-09-05 Protein FF meeting note

MBAR analysis

Open

@Chapin Cavender

• MS – I’d have thought it’d be faster than 4 weeks… is the bottleneck running new sims?

  • CC – Yes

  • MS – So optimizing to the number of effective sample should be quick, but you’ll occasionally need to collect more samples?

  • CC – Yes

  • MS – I’d think it’ll converge faster than that.

  • CC – Optimistically

  • MS – I’m thinking a smaller energy is likely… (ff14sb showed 20kcal/mol when folding from 0.4 native contacts)

  • MG – We’re not fitting against free energies, we’re fitting against NMR

  • MS – Makes sense.

  • CC – Even a less steep

• LW – If you use OPC3, do you anticipate running into issues with params optimized against TIP3P?

  • CC – MG and I talked about this a bit. So my answer is yes, but my concern is that TIP3P makes proteins too compact. Also, we’ve only trained against TIP3P for small mols with HFEs. I think our lack of floppy mols in training against TIP3P means that TIP3P and OPC3 are

  • MS – Should train against TIP3P?

  • MS – We’re finding that OpenFF performance against TIP3P-FB and OPC3 is ok? … (see recording ~25 mins). Might ask gov board.

  • MG – What kinda concerns me is that sage LJ is adjusted to work with TIP3P. So in a sense we’re mixing water models.

  • DM – We did try the LJ params that were more AMBER-ish. Our top priority is getting a protein FF, and refitting to another water model is a minor issue there.

  • MS – Possible that we’d need to fit protein FF to water model, then refit small mol, then refit protein?

  • MG – would using TIP3P prevent this from getting done? Or do we expect one would be easier?

  • DM – I’m saying you should do what you think would be easier, and we can refit the general FF to the water model you pick.

  • CC – I think it’ll be easier to get our energy vs. native contacts plot looking correct using OPC3.

  • MG –

  • MS – If we do OPC3

  • LW – I was going to suggest that, on this 4 week timeline, one thing we could try doing during that time is refitting Sage using OPC3. I guess this wouldn’t including the fit to protein QM data, how long would that take?

  • CC – To go from a QM-fit FF to an NMR-fit FF is about 4-6 weeks, regardless of starting point. So if I could take this FF that I’ve already done QM firs on (null or specific 0.0.3), and then meanwhile you refit as described, we can run BOTH FFs through NMR fitting. Then we could see whether the effect of water model on the starting point.

  • LW – I can do this.

  • MS – Was wondering if there was some overlap with BMorales work where she should help LW

    • LW – Maybe further down the line.

  • LW – CC, what’s your starting FF?

    • CC – Sage 2.1

    • LW – Ok, I can start from that too.

    • MS+MG – Why not use sage-2.2?

    • CC – The targered chemistries for 2.2 didn’t include proteion groups

    • MS – But nonbondeds were retrained?

    • LW – Yes, but I assume effects will be slight

    • CC – If we train both protein FFs (2.1 + protein QM fitting…

  • MG – Does this affect governing board question for monday?

    • DM – I don’t think so. The question being asked is “when do we release a workflow for handling PTMs using ff14SB+Sage?” - The thing we want to determine there is a deadline for when we go for that vs. waiting for protein FF.

• JW – I like the validation step - Very clever. How will you determine if validation passes?

  • CC – Want chi2 to be within 1 of real ff14sb

•  

  •  

•  

•  

•  

•  

GB3 NMR torsion fit

@Chapin Cavender