2021-02-19 WBO/Impropers meeting notes

Owned by Pavan Behara
Last updated: Feb 15, 2022 by Diego Nolasco (Deactivated)
3 min read

Participants

@Jessica Maat (Deactivated)

@Christopher Bayly

@Trevor Gokey

@Simon Boothroyd

@David Mobley

@Pavan Behara

 

Discussion

JM: Here is an interactive graph that shows the difference in wbo values from openeye(OE) and ambertools(AT) packages. The code resides here.

CB: Looking at the outliers on the right hand side OE says single bond and AT says aromatic bond. There should be difference in geometries due to optimization under the hood.

SB: Yeah, the initial geometry is perpendicular and AT makes it flat whereas OE keeps it perpendicular.

CB: I think my restraints are over-restraining in this case. They’re keeping the bad geometry, the intent was to not change the dihedral too much. Omega looks at the molecule graph and says it should be perpendicular, AT lets it go flat as much as chlorine allows electrostatic collapse.

SB: I found another instance where AT allows proton transfer, this is a thin line to walk.

CB: The question here is how much does it affect wbo?

DLM: Does harmonizing restraints a solution?

SB: It is still unclear, we can do restrained opt in AT same as OE way but that means we are sort of reverse engineering OE methodology.

CB: Overall optimizing the electronic structure is needed. Is there some science where we can do an optimization and figure out if we have an electrostatic collapse then we apply restraints o/w we stick to full optimization? Can we do it post AM1 calculation see if it is bad and then go back, can we do it before Sage?

DLM: We do need consistent wbos from both the packages for sage.

CB: Main issue is electrostatic collapse, what if OE way of doing things is wrong?

SB: We can do some quick & dirty checks on electrostatic collapse but we will still have issues.

CB: What if we build a FF for AT approach which has occasional catastrophe but better wbo & OE tools separately.

DLM: We can bring OE in line with AT

CB: So we were very cautious to avoid the catastrophe that resulted in the use of these restraints, bad results are fine but never ever a catastrophe.

DLM: Can we do this SB?

SB: Let me think it over, we may have to benchmark heavily.

CB: This is an engineering problem, we just want a good wbo.

DLM: The other thing we will get something working with AT & we will wait for OE to keep working on it.

CB: There is room for catastrophe.

DLM: We have to make AT avoid the catastrophe.

SB: May be looking at the changes in topology, or other options.

CB: I may have a pseudo code in my mind, what if for AM1

  • we begin the OE way and get charges

  • AT way get charges

  • do a comparison and if the difference is beyond a set threshold then we fall back to using restraints the OE way o/w we will do the full optimization

  • all these steps under the hood

DLM: SB is that reasonable?

SB: I will think it over

DLM/CB: Thanks JM for showing these plots, this is really helpful.

 

PB: Here are the plots for TIG0 after splitting into two parameters, one with carbon only atoms on the dihedral, the other with heteroatoms allowed on one end.

CB/DLM: They’re the same molecule and yet the torsion profile is different with the same interpolated FF.

PB: Let me check the molecules, I may have pasted a wrong picture, nope they’re the same but from different datasets. No idea, I might have made a mistake somewhere.

TG: Check the canonical smiles, sometimes kekulization might be different.

SB: Nahh, for 1.3.0 the torsiondrive is same and even QM energies are matching.

PB: My apologies, I will double check this and report back.

PB: I also looked into how Chaya’s trends look with the barriers from residuals, here are the plots that show both QM and barriers from residuals calculated using 1.3.0. The barriers are higher but looks like linear trend holds in most cases.

SB: That’s wishful thinking, there certainly is lot of noise, I also looked at the same thing today.

CB: Yeah, I agree with SB, there is lot of noise and we are interpreting the way we want to interpret. I would still like to see how the splitting works.

SB: I am looking at JM’s newly submitted datasets, it’s still going on, from my analysis I could see TIG2 shows a good linear trend and rest all are noisy.

DLM: Impressive, the range of wbo’s looks good.

SB: Yeah, may be we can focus on this more.

DLM/CB/SB: Thanks to JM & TG for creating these clean datasets.