2024-12-12 Protein FF meeting note

GB3 NMR fits

Open

@Chapin Cavender

• SLIDES & RECORDING

• slide 2

  • MS: So you ran the simulations with blue and you try to run predicitions using the blue data

  • MG: He was trying to move the probabilities towards zero. The dashed lines were done posthoc.

  • MS: Blue line is running the simulation, solid orange is another simulation and the dashed orange is trying to predict from blue using the orange energy function. # of effective samples?

  • CC: # of effective samples is tens of thousands.

  • MS: Something might be wrong here, I can take a look at the code.

  • MS: If you have 10’s of thousands of …

  • …(some complicated discussion of data on slide 2 and methods on slide 24, see recording up to 10 mins)

  • CC will upload code/send to MS for further debugging

• slide 7

  • JW: trajs are newly run for iteration of FF fitting, or same three trajs are reweighted for each FF?

  • CC: Newly run for each FF iteration

• slide 9

  • JW – You might draw a y=0 line across the plot, to show which FFs have a favorable folding energy

• Slide 13

  • MS – How stable is this alpha helix supposed to be experimentally?

    • CC – Unknown, we only have NMR data (slide 10) which indicates it should be more stable

    • MS – So ff14SB DOES match expt well by keeping things very folded

• slide 15

  • MS – Significant that the Neff for the 1E4 series is so low

  • MG – … (see recording ~24 mins, something about tradeoffs in sampling)

• CC – MS, how could I check that, when using cumulative inputs from fits, I’m doing it right?

  • MS – Check dimensions of matrix to ensure correct size, and that chunks from different FFs have the numerical differences you expect. Maybe plot distribution of differences from first group of samples (from FF 1) since you’d expect significant changes from FF2, etc.

  • MS – Would like to understand what’s going on with butane example. But it seems subtle.

  • CC – I’ll send over code I’m using for butane example.

• MS – Should we do additional QM datasets?

  • AF – Currently trying to determine the best way to select conformers. Working with CC on this, might write up a proposal and send in Slack. Or could present at next meeting.

  • JW – Probably presenting in a meeting would be good for interactive discussion, but this wouldn’t preclude posting in Slack

  • AF – I’ll post a writeup in slack and we can comment async.

  •  

• JW: Looking at the free energy funnels I feel the yellow one looks like a good candidate, convince me that it’s not?

  • CC: chi2 shows it’s not as good as ff14sb

  • MS –

  • MG –

  • JW - What if GB3 is a rigid protein and ff14sb being rigid performs really well

  • MS – AF could try other proteins

  • CC – AF has simulated before with ff14SB, could do this

  • AF – We have compute time available.

• MG – Is there any pattern you see in the parameter changes?

  • CC – Biggest numerical change (both abs and rel) is general psi parameter. In ramachandran plot alpha and beta have roughtly the same PHI value, so PSI is the thing that differentiates them mroe.

  • MG – What if we put our thumb on the scale and pushed psi even further (even just directly modifying numbers in the FF file)

  • CC – Could do that. And could evaluate quickly on a short dipeptide

  • MG – Right, and we could quickly re-check the chi2 in these experiments.

  •  

•  

•