Results of different fitting strategies | JH – Tried using (1) all torsion terms available, (2) just parsley torsion terms. JH – Is there any downside to only doing ONE bespoke parameter on top of Parsley? Like, if there were several torsions running through a central bond, could we bespoke-fit just one of them? DM – Trying to think of cases where things on different sides of a central bond AREN’T as rigid as aromatic rings, and might need to know about their flexibility. We wouldn’t want to flex certain parts of end groups. JH – I’m assuming that they’re relatively rigid SB – Could we make SMARTS where it’s the entire molecule SMILES, with the 2- and 3- atoms tagged explicitly, and any atom that is in the 1- or 4- positions would be written as a wildcard. DM – Not sure that this would work, this might break something in the construction of SMARTS.
JH – Found some cases where we do poorly relative to GAFF. Refitting
(?starting from GAFF?) could fix these. Where would be the best place to report these? DM – Striketeam channel. Trevor could use this info. JH – I use a higher energy limit before downweighting, might be important to mention as well. Will also try running with different priors.
JW – If fitting multiple torsions would produce a certain energy profile, are we sure that a single torsion would reproduce the same energy profile? Or would it be necessary to include many to encode different phases? (General) – This isn’t mathematically proven, but it seems likely based on our experiments that fitting a single torsion is “good enough” DC – Case like dibromoethane would trigger this sort of behavior. JH – Right, I had to split that into multiple terms.
DM – If we come up with code to generate a single torsion SMIRKS that “fits everything” for a particular rotatable bond, we should share that with JM.
|