LibraryCharge ESPs across amino acids | LW | Librarycharge ESPs on peptide chains CBy + JW – Why do caps have such high ESP magnitude? (they’re not net charged). PB – Is chaping using librarycharges in his protein fits? JW – So it looks like we have the ability to polarize our termini if we want. Do we want to do that? CBy – The history of caps is that they let us cut off wiggly bits of proteins (termini or loops) so that things can converge. Also for simulations of short peptides (but I don’t know whether this is actually done in pharmaceuticals). Another piece of lore is that caps are on parts of the protein that you expect to be exposed to solvent. But if we want things to be “realistic” I don’t see why caps would be more “realistic” than a typical amide polarization that you’d find eg. mid chain. JW – Given that CC is having trouble with short-ish peptides, I think it could be good for him to try this out. CBy – I suspect the major difference in ESP are due to differences in the dipoles here. MS – Does this mean we should use Nagl instead of librarycharges for proteins? LW – I think it’d be worth looking at how the cap librarycharges were generated, since they were made using ELF10 but in my work I’m seeing the caps being really different vs ELF10. CBy – I’d also think a torsion refit using nagl charges could lead to substantial differences to torsion params. I’d expect that to have a significant effect on benchmarks, especially around caps. CBy – Let’s remember that we mostly care about AAs in the middle of proteins. Does NAGL get those right? LW – Yes, generally NAGL is quite close to librarycharges for mid-chain residues. Good to keep in mind that chapin generated his librarycharges using fairly small mols. JW – I think it’s hard to reason around what we’d expect comparing chapins librarycharges vs. nagl charges, given that neither one has yet made a successful FF. So I think the only way forward here is to put things into action and see if chapin wants to test something.
|
Free energy comparisons with NAGLrc2 | LW | MS – Surprising that NAGL charges are closer to OE ELF10 than AT, but the RMSE is worse than both. DM – Given that these results are for freesolv, I expect higher error/more excitement on a more diverse set. MS – FEP could narrow these errors CBy – Did you see any systematic differences by functional group? DM – This is interesting, MO in my group is looking at charge differences and how they affect fe calcs. She’s had trouble finding things that correlate with error, so I’m amendable to the idea that maybe these hydrogens on the hydroxyl carbon having a slightly different charge could lead to significant differences. CBy – Is NAGL trained to point charges or ESPs? LW – Both CBy – Even if we find out what the cause it, what can we do with that? DM – I think this shows … and that ambertools charge assignment has a lot more conformation dependence. CBy – That could be, the logic of ELF10 is to reduce conformational dependence. So if AT were using a difference conformaiton, it could just as easily be UNDERpolarized rather than OVER as we’re seeing here. LW – I think that we acknowledge the conformational dependence of these methods. What we’re showing here is that the conformational dependence can lead to changes in FE calcs CBy – With nonaq solv FEs, we’d expect inaccuracy (with a fixed charge model that’s overpolarized) is when the solvent is nonpolar. How is the accuracy for solv FE calcs in nonpolar solvents?
|
