2020-04-23 Torsion Interpolation Meeting notes

Implementation plan

@Jeffrey Wagner and or @David Dotson

• Overview of the current plan

• Where do “curves” come from?

  • Fitting process will encode them in each force field

• CS – Previous work has been restricted to only refitting k terms for terms with periodicity=2. Do we want to do the same?

  • DM – We can experiment with this.

  • JW – If we have multiple Ks for the torsion, but only want to vary one K with bondorder, but implementation requires the same number of kXbondorderB for each, then we can just provide the same K for each bondorder in the “vanilla” torsions

  • DM – Would be helpful to identify key functional groups/molecules to look at by modifying notes after meeting.

• CS – Are we looking at doing bond/angle interpolation as well?

  • (General) – Yes, but after torsions

• CS – I’ve found some molecules that would be really good for tests, such as substituted biphenyls

  • JW – We should separate benchmarks and unit tests

  • (General) – We can include a few of these in the toolkit’s integration tests

• DD – Is current implementation sufficient?

  • JH – May be useful to have per-torsion interpolation scheme? Would it be OK to do “linear” for an entire section, or should we allow each parameter to have a different interpolation scheme?

  • (General) no – this would just add another fitting parameter. We can add it in a year or two if needed.

Working with several torsion types per torsion scan.

@Chaya Stern (Deactivated)

• Weighting different torsion types to get the asymmetry for chiral torsions

  • Could allow phase angle to be other than 0/180

  • Could allow different “weighting” of the many torsions, so that cosines don’t combine to be symmetric

  • JW – If a nitrogen gets conjugated as a function of torsion angle, then this may be an inherent problem with our cheminformatics representations

  • DM – Two thoughts:

    • WBOs let us avoid some problems with cheminformatics representations

    • I don’t want to represent parameters as chiral

  • CS – If a torsion involves a trivalent nitrogen, the ones with lower WBOs do not conjugate. It may be possible to represent this in the FF.

  • JM – So, would we calculate WBO around trivalent N, and interpolate differently based on the behavior we see?

  • CS – You’ll get the same improper torsion within different environments. So analyzing central bond character of connected proper torsions may be informative

  • CS – If nitrogen DOESN’T conjugate, you get asymmetrical PROPER torsion scan curves.

  • CS – If you look at a scan of an IMPROPER torsion, it may be possible to find a threshold where it favors remaining pyramidal vs. planar.
    JM – Would be great to have molecules that display this behavior

  • JM – Data for biphenyls in in QCA?

  • CS – Yes, this is in a torsiondrive dataset (CS will paste name/link here)

• In the above, two questions were discussed:

  • For a PROPER torsion where the 1- or 4- atoms are a pyrimidal nitrogen center (or other chiral center maybe?), the torsion profile is expected to be ASYMMETRIC – How do we represent this in the FF?

  • For an IMPROPER torsion, where the center is ALSO a 2- or 3- atom in a PROPER torsion, sometimes a pyrimidal center will favor being pyrimidal initially, but for certain angles of the PROPER torsion, the pyrimidal center could favor being planar.

• DM – Be cautious about encoding chirality in SMARTS – This will impose contraints on the generality of training data

• CS – I need someone who will take over fitting to/owning biphenyl set. Otherwise I’ll go ahead and try to publish it in its current state.

  • JM – I’m potentially interested. I’ll look over the current data and plan. Will contact Chaya to discuss.

  • CS – My timeframe on this is that I’m starting a new job in mid- to late-may, so we should make a decision before then.

Trivalent nitrogen

@Chaya Stern (Deactivated) and @Jessica Maat (Deactivated)

• Discuss coupling of torsions with trivalent nitrogen.

  • (Notes for this are mixed into the above section)