2023-01-12 Protein FF meeting note

LiveCoMS review

@Chapin Cavender

• CC – MG and I have submitted a major revision of the paper to coauthors. Expecting this to be a final push, we’ll be tracking people donw to get back to us. Ideally will be finished by the end of the month.

  • MS – That’s great timing. This will be a big help to the R01, especially points around building community.

QC validation for protein force fields

@Chapin Cavender

CC will upload slides here

• Sage-PB: ~2.1.0

• Protein-Null: starting point for protein null model, without protein-specific parameters

• …

• JW – “22 of 54 torsiondrives” – Should we prioritize fixing this?

  • CC – I don’t think it needs to be fixed for the NIH submission, but it would be nice to fix it by the time we publish. Right now the sets are marked as needing scientific review.

• LW – Do the parameter values in protein null end up close to the protein-specific values?

  • CC – Big difference ends up being wholly different periodicities.

  •  

• (JW has a lot of back and forth saying that the protein-specific FF seems to be doing the same or better in validation plots, but CC reaffirms that there are cases where protein-null does better and points out that the overall summary plot shows they’re about the same)

• LW (slide 18) – It’s weird that the protein-specific FF doesn’t have a dip at +-180. It should be the one OpenFF force field that has the torsion periodicities to put a dip there to match QM/ff14SB

• MS – What’s the plan for running the NMR validation, and will it use protien-null and protein-specific?

  • CC – It’s a tiered approach - 1) short peptides with null and specific FFs, then whichever works better will advance. Then tier 2 is larger folded proteins. I’ll run both tiers with ff14sb for comparison.

•  

Benchmarking using protein crystal structures

@Michael Shirts

• I have a student that has the background and interest in doing this. Looks like there are resources. Is this decent priority (student is very good at running and analyzing biomolecular simulations, not as much experience with software development).

• MS – If sims are big and we’re looking for compute time, would it make sense to test things out in gromacs (compare FFs using gromacs instead of openmm)? Or are there better things for her to work on?

  • CC – That could make sense. Are you worried that something might not be supported in OpenMM/size could become an issue?

  • JW – I think the plan was for chapin to run part of the dataset on the UCSD cluster and for Lily to run the rest on lilac.

  • CC – From the livecoms review, there seems to be consensus on how to validate on NMR, but there aren’t really standard workflows for crystals.

    • Basically, there wasn’t consenbsus on how to run sims of xtals and then run an analysis to compare to observables. One idea was to make an electron density of the sims… (some subtleties here, article probably covers them better)

    • CC – I’ll post a link to the article.

    • DM – It seems like it could be done, but it would take a few postdoc-months. Seems like a lot of the metrics are manual/done by looking at it.

    • MG – In D3R, there were discussions about comparing to electron density for docking (rssr or something). It was highly opinionated and different metrics had problems with sensitivity.

    • CC – It’s an interesting area that probably would require a bit of work. I’m not prioritizing that for the protein stuff because it’s more straightforward to do NMR. But we do have the datasets available.

    • MS – Sounds good, I’ll talk to the other PIs about future plans/allocations

  • MS – If it’s a valuable dataset, then that may be worth testing. Could you pass along the LiveCOMS review and dataset?

  • CC – Sure – I’ll pass the dataset along.

  •  

•  

•