GB3 benchmark simulations | View file |
|---|
| name | 2023-04-06-dihedral-hists.pdf |
|---|
|
| Chapin Cavender | CC will link slides here JW – So, slide 37 is showing there’s a kinda barrier “added” in ff14sb that might prevent transitions over phi=180? MS – Does it look like we have stabler alpha helices? … – Hard to tell, more analysis needed. MG – It’s possible that our hbonding energies are different. Could do gas phase calcs to see
“slide 52" JW – Hard to tell what’s up here - There’s so little alpha helix propensity in the 3-mers that it’s difficult to figure out MG – CC – Could study sidechain rotamers more , instead of just phi and psi CC – Could also look at tau angle - Some xtal structure info on that MG – If you look at GB sims where helix partly unfolds, is there an initiating event that we can identify? Like the helix pulling away from the backbone or something? CC – I don’t see anything by eye, but I’m thinking of checking out the tau or omega angles. JW – Maybe something with ions? CC – Using the same ion params for all, there’s just 2ish ions in the box to match experiment conditions
JW – Also re: previous comments that maybe our hbonds are too weak, we’re seeing that beta sheets are increasing while we’re losing alpha helices CC – CC – What additional experiments should I prioritize with my time?Currently first priority is alpha helical sims, the sidechain analysys for existing sims, then hbonds in existing sims and maybe in gas phase dimers, and priority 4 is tau angle. MS – This is a good proof that the benchmarks suggested in the paper are good PB – How are the force constants here? CC – Some periodicities have ks over 5kcal/mol, but that’s also true for ff14sb PB – The impropers might be a contributor to the error here CC – Do you think the impropers might have k values that are too large? PB – Yeah, maybe. One improper changed a lot from the previous values we were using - (i4, not changed in openff-2.0.0, but refit in Chapin’s starting point FF)? JW – The nitrogen centered improper may be i6?
MS – This has shown that, even if we get everything right that we intend to get right, it might go wrong for other reasons. So this could be a good reason to expand our validation set… replica exchange? CC – Maybe melting temps/helix propensity vs. temp? MS – Given how GB3 didn’t work even though QM fits looked good, J couplings look right, we might need a larger validation set. This will head off our future users reporting issues after release. CC – That’s how I’m thinking about this as well - Would like to get to the bottom of GB3, and find a way to assess that sort of problem more directly.
JW – Could the energy surface difference be due to different QM reference data? CC – Amber used MP2 (which is widely considered to do dispersion correcitons better than our basis+method). But IIRC PB found that our results are very close to MP2 with dispersion. PB – Yeah, my analysis found that we’re very comparable in terms of QM methods. CC – Notably, in ff19sb, they used a DFT model in implicit solvent. MS – So there may be some differences, but it seems likely.
CC – There’s also a difference in how the fitting is done - CSimmerling fits to conformers in a different way than we do, which is kinda what Swope suggested. But our fitting infrastructure would need fundamental changes to do this.
|