2024-10-17 Protein FF meeting note

GB3 NMR parameter fits

Open

• CC: will create a MWE of the issues between OpenMM energy and reweighting potential. If this doesn’t diagnose the issue, will post on Slack and ask to screen share with someone from the infrastructure team to fix this.

• MS: results from slides 3-5 show one just needs to be careful about equilibration

  • CC: Yes, probably what’s happening is the initial window is generated via steered MD. Seems reasonable to expect that there are fewer configurations that correspond to the native vs non-native structure, so less-folded windows may take longer to equilibrate.

  • MS: seems reasonable – eventually, replica exchange would help, but seems unnecessary now and would introduce complications

  • MS: Also, it’s getting the direction right so seems great.

• Slide 8:

  • MG: possible the odd results are due to the bug above.

  • CC: hoping that resolving it will make this behave better. Think we’re close enough that we’re getting some signal out of this.

  • MS (row with NMR 1E2): I’m surprised you have a high number of effective samples (196e3) and got a chi2 of 11.8. What’s the total sample?

  • CC: total pool would be 15000 x 31.

  • MS: that mathematically almost doesn’t make sense. Where did the samples that gave 2.8 go? That looks weird.

  • CC: possibly more evidence that the energy function isn’t correct

• MS: will you need to re-run simulations again, or just re-analyse?

  • CC: not the initial simulations, but the later ones…

  • MG: if the reweighting script generated parameters that don’t move us in the right direction, we should rerun those

• CC: I’m at a point where once an umbrella sampling run finishes, I can analyse the fit in a couple hours. I can use UCSD resources to run a fit in days. Our goal was to run multiple values of alpha.

• (recording around 25 min – clarification from PB on potential functions)

• MS: points out a number of issues around torsions

• MS: happy to take a look at things – spent a long time matching dihedral energies for InterMol

• MG: three failure points

  • In script that computes the quantity

  • In the parameter file that OpenMM reads

    • CC: I’m using OpenFF Toolkit to write out new parameters

  • Reading in parameters into OpenMM. Seems unlikely

• (some discussion of possible issues, recording ~30 min)