(Slide 9) JW - What region of the left plot (heavy atom RMSD) do we care about? I think we don’t care about RMSD < 0.3 Å, and we care a lot about RMSD > 0.6 Å.
DM – Worried that the high RMSD region is dominated by things that minimize to somewhere entirely different.
LW – Kinda agree. And 0.3 Å RMSD captures cases like planar rings that MM models as buckled. So that’s where I spend a lot of time looking.
JW - Re: DM’s point, isn’t minimizing to a different conformer exactly what we want to avoid?
DM – But you’re starting an optimization somewhere, and a slight difference in gradients between QM and MM could walk things away to different minima
JW - But these optimizations are started from QM minima, so I wouldn’t expect that to happen
DM – Big tail has two sorts of things:
Taking a good geometry and walking far away from it due to problems with MM landscape
The MM optimization fails to see it is in one minimum, but goes to a different valid QM minimum.
LW - We do see examples of electrostatic collapse – one example I looked at had RMSD > 10 Å, so those conformers aren’t even on this plot
JW – Agree that these are problems with MM FFs. But if I had to make an algorithm to look at this plot and determine which FF is “best” I’d look at the order of curves at x=0.8 or so.
DM –
LW – Given that we have multiple sets of minima, I could check to see whether the ones with high RMSD are just finding another minimum.
DM – Could look at AUC as a useful metric
Video Conferencing, Web Conferencing, Webinars, Screen Sharing 