CC – “Diverse selection” slide, how is diversity defined?
CB – “electrostatic diversity” → Basically each atom gets an electrostatic energy (energy of it interacting with all the other atoms). Then this becomes a feature vector that I can use to define diversity. I use “Gonzala’s algorithm” to select the most diverse of these feature vectors
JW – How much are MMFF charges a limitation?
CB – I only checked MMFF. The goal of this initial charge method is to RANK the conformers, not to get their energies NUMERICALLY right. So this probably isn’t the big limitation.
CB: MD constrained, are bonds, valence angles changing/relaxing?
CD: Yes.
Discussion topics
1. We want to stop proton rearrangements
2. when there are no proton rearrangements OE opt may be harmful, how bad is it?
3. We’re making a restraint scheme - What should our “gold standard” be for “doing it right”?
CB:
One universe says “there is a best set of AM1 charges for a certain geometry, and they are conformation dependent”. I don’t like this because strongly internally interacting structures make a mess of everything – Even when the conformers extend, they get stuck to water or protein contacts.
Another universe says “One single set of charges is the best for every possible conformer of the molecule”.
The evidence seems to indicate that we want the second option. So, the approach in part two of the talk (where you consider AmberTools AM1 as the “metric of success”) is only applicable to universe 1, and I don’t think that's the best way.
Solvation energies and relative energies of conformers are good metrics of success
So the thing you need to answer is “are really strong internal interactions allowable?” and I think that they’re not, since universe 2 is more important.
If we use a more detailed QM method as a metric of success, then we need a way to keep those from having internal hbond rearrangement/strong interactions.
You want a “redundant internal coordinate optimizer”, optimizing bonds and angles, but (con/re)strain torsions
SB – GeomeTRIC offers this.
CB – What really seems ideal is to let everything relax, as long as we don’t get “electrostatic collapse”, where we get strong internal electrostatics and hbonds.
CB – If you’re gonna use a fixed charge FF, you have to answer what the metric of success is. I used relative conformer energies and solvation free energies.
JW – Re: Comparing to conformer energies, we noticed that anywhere that AM1 would have had a proton transfer, then high-level QM would also have a proton transfer, How should we handle that?
CB – The real problem isn’t proton transfer, it’s electrostatic collapse. That should be where the failure is considered to happen. So IF you consider this to be a failure, then this should be removed from the dataset altogether, or you should find some other way to “repel” away from those collapses.
JW – So, in conclusion it seems like, universe 2 is intuitively more fit for our needs than universe 1. Basically, we can’t define truth any better at this point without running high-level QM or hydration free energies. We also need to define whether electrostatic collapse is OK, or what it is.
JW – Case of antechamber vs. quacpac disagreement – Should we report to OE?
CB – It’s probably AM1 converging to a different electron distribution. But this is probably a super hard-to-fix thing, and it’s not clear whether it’s in our code or ambertools. Also this is one molecule out of 500, and we don’t know whether the downstream effects of this difference leads to a meaningful difference in relative energies.
CB – I outlined a restraint scheme to Simon a few months ago. It was basically “run a restrained opt, then run a full opt from the outcome of the restrained opt. Then see if the full opt has an electrostatic collapse, go back to the results of the restrained opt”
it would change torsions and everything (would improve)
Direct polarizability
response of polar fragments in non-polar envs (such as interior of a protein and lipid membranes)
right now dielectric of lipids is constant
SB:
WBOs
We’re continuously evaluating WBOs in our fits to establish when they’re ready to get in.
VSites
We’ve got the infrastructure to fit vsites. The cole group is working on this, and TGokey. So there are feasibility studies ongoing.
Polarizability
Gilson group started looking into this, I was hoping that someone there would pick it up and run the feasibility study on this. I’m not sure about the status of this. So we may be blocked by personnel on that.
DM – We also don’t know exactly what we want to support in the spec for that.
CC – MG has a master’s student (Willa Wang) looking at polarizability.
CBy – I’m meeting with Mike and Willa next week. So my question is “what would this take?” – Certainly proof of concept, a spec that’s forward-thinking for interoperability. So I’d like to get to a go/no-go point on this.
DM – It’s largely a personnel problem - We don’t have a person to do the research cycles and infrastructure work.
DM – I wasn’t aware that OE was working on xtal stuff. If it’s an area of major overlap we could host a co-employee
CBy – That could be a heavy lift, it’s hard to get support for this sort of thing at OE.
SB – Shirts lab is bringing on a person to do xtal stuff. So there could be some support on proofs of concept there.
CBy – May be a conflict because we’re so focused on small molecules and biopolymers. Xtal structure prediction may not be aligned enough to be a win for us. So it would be of greater interest to us to have a good general FF for biological interactions. I’d focus on filling in evidence and infrastructure gaps for polarizability. So I’d like a roadmap toward integration of polarizability.
SB – We need an infrastructure and science proof on concept on this, likely from the Gilson group. Once that’s in place, then we can schedule infrastructure and data work.
Action items
Decisions
No labels
0 Comments
You are not logged in. Any changes you make will be marked as anonymous. You may want to Log In if you already have an account.
0 Comments