| CD + JW – We’re finding that OpenEye AM1 charges aren’t actually taht good (they often seem to have proton migrations). Thing that we’re looking for is a dataset of RESP charges. TG – Could use RESPYTE, or AmberTools RESP tools. JW – Where are your current datasets with partial charges coming from? TG – Right now I use the OpenFF toolkit, and basically treat them as toy charges. CD – Here’s why we think OE is having proton migrations.
JW – So, we’re going to be trying out alternatives to “maxcyc 0”, but we want to be able to determine which restraint schemes TG – There’s only one dataset in QCA that has wavefunctions – That’s the BCC refit dataset. We’d need to take the wavefunctions and run RESP. But do you really want to fit to low-level QM-RESP charges? (General) – What are we trying to fit? If a molecule wants to rearrange in vacuum, what is its “correct geometry”? If it’s only stable in solution, how can we model that in our workflow? TG – Maybe it’d be better to focus on finding good charges for a representative set of conformers? Or to trigger a search for a different conformer with less electrostatic interactions? JW – This could dovetail into the ELF implementation JW – Seems like there are two questions: Is there a restraint scheme that we can ALWAYS apply, and that will give “good answers” (this would require a “good dataset”) What if we don’t restrain things, and we detect a proton transfer. What restraint scheme should be we use there? How would we validate this restraint scheme?
CD – Are conditional restraints bad for users? TG – Conditional restraints are bad because they depend on the input conformer. CD – What if they’re only applied if a proton migration happened? TG – That wouldn’t be as bad. But what if you just found a different geometry?
TG – I’d recommend also looking at the RMS and the max deviation.
|
