2022-08-11 Protein FF meeting note

Starting point for protein FF

@Chapin Cavender

• CC would start generating protein parameters

• CC: At what point can I merge the improvements from small molecule FFs into my current work?

• BS: When the first version of biopolymer FF comes out does it have to be supplemented by a small molecule FF?

  • CC: Expectation would be that parameters generated to be used at the protein ff would also be useful to a small molecules ff.

  • JW: At the gv people raised the possibility of a new Sage release. There was concern that the person-hours required will push back other deadlines, so we are still discussing this. Rosemary would be a single ff for proteins and small molecules.

• (General) – What should CC’s project use as a starting point for turning a small molecule FF into a protein FF? This is important because the “starting parameters” in the small molecule FF may be insifficient to describe protein sidechain torsions, so we know that we need to make protein-specific sidechain torsions. But if we make this decision based on one Sage 2.0.0, but then the state of the art is Sage 2.1.0 when we need to make Rosemary, then those decisions may no longer apply.

  • JW - Start with Sage 2.0.0 so that you aren’t waiting for the decision about a point release

  • DM - Start with Pavan’s, since those changes are almost certain to get into a future Sage point release of Rosemary release

  • LW - Agree with DM.

  • MG – Maybe we can hedge against the uncertainty about exactly which parameter splits will happen/which torsion types there will be by defining a PROCESS to decide whether the protein FF adds a special term.

  • CC – I think this decision tree could get really complex. I don’t know how we’d handle all possibilities

    • Decisions like whether an amino acid is charged or not, a backbone torsion is better or not, …

    • MG: You’re saying whether we can carry over small molecule FF parameters that do well for proteins as well?

    • CC: Yes.

    • MG: I think this would be less of a problem with accuracy but more with parsimony. We can look and see if there are any torsions that can be merged.

    • DM: Only worry would be pace of work.

  • CC – Is Pavan’s current work far enough along to be the best starting point for me?

    • DM – Yes

    • Decision – CC will begin making the protein FF starting at Pavan’s latest FF candidate.

      • CC – Agree

      • MG – Agree

  •  

•  

Protein library charges

Slides

Open

Protein library charges

@Chapin Cavender

•  

• Slide 3+4

  • MG – NH charge has a big difference. Same with CA

• backbone Ca differs a lot with ELF10, I think it is slightly polluted by the RESP fitting. We should not have more polar NHs so we should look into these carefully.

• CC: I think the NMR observables would help us resolve this, the secondary structures

• MG: I don’t know if we benchmark this with crystal data and solid data but something to think about.

• CC: Yeah, crystal observables are tough to set up so it is on my second tier of benchmarks.

• MG: Yeah, Tobias might help here.

• CC: Yeah, once I have a FF I can check in with him.

• Slide 18

  • DM: To some extent RESP is created to account for buried charges (on carbons) and that’s why we are seeing more polar or larger charges that are unreasonable. I saw them before as well.

  • CC: I agree.

  • DM: I think that’s why we don’t see those larger charges with AM1 BCC.

• BS: I am intrigued by amide carbonyls and carboxyl carbonyls having larger charges, I am concerned about helical

• • CC – Full twist of an alpha helix would have 7 residues. We wouldn’t be able to train against things this large - We’d just be able to test them in NMR sims of peptides.

  • MG – Do CHARMM FF residues have larger/smaller magnitude chages?

  • CC – I think CHARMM charges are also smaller in magnitude.

• CC: CHARMM would like to have integer charge and AMBER would like to redistribute

• CC – The benefit of having the same charges as the small molecule FF probably outweigh possible biases in secondary structure

  • MG – Maybe.

  • BS – A few years ago, in the charmm FFs, people were seeing too much secondary structure. Then they did something to the FF to fix it. Do you recall what that was?

  • MG – Maybe the added CMAPs or reduced charges?

  • JW: So, if we have to catch this in training we are gonna need a full alpha helix and beta sheet and CC you say that we don’t have any such data? If it is to be fixed by torsions in other FFs should we have a plan of action for that?

  • CC – 1) Sage LJ terms may help us resolve this - They may just be better and so we’d avoid these problems altogether. But 2) if we DO see this problem, I don’t think that larger QM datasets will help. So we’d need to fit LJ types for proteins.

  • MG – There may be a bigger question here - If AM1BCC charges are good for small mols but bad for ptoeins, what does that mean? Is AM1BCC only suitable for solvent-exposed small molecules, but fails in buried situations? Then we may need a different functional form.

  • DM – I have some confidence in AM1BCC - I think we may not hit the right secondary structure balance but there will be a lot of other benefits. Do you have advice, BS?

  • BS – It would be good to ensure that the training set contains energies of alpha helices and beta sheets.

  • CC – We’ll have

  • JW - The NMR data is used in testing and not in training so it would diagnose a problem but it won’t solve it?

  • CC: Small 1st tier benchmarks and choose the winner to go for the 2nd tier larger protein benchmarks.

  •  

•