2020-11-12 Torsion interpolation Meeting notes
Date
Nov 12, 2020
Participants
@Jeffrey Wagner
@Jessica Maat (Deactivated)
@Pavan Behara
@Hyesu Jang
@Simon Boothroyd
@Joshua Horton
@Trevor Gokey
Getting “torsional energies”:
HJ + SB
After adding torsion-specific parameters for the molecule to be fit, let them have multiple k values
Fit those new torsion-specific parameters
Make a new OpenMM System based on Parsley + the new torsion-specific parameters
Put those new torsions in a separate ForceGroup, and recalculate conformer energies
SB – Look at torsions and identify those with the correct central bond
TG – Zero out all other torsions
What we’re trying to achieve
DLM's comments on lab slack: "there are two groups experiments we’re talking about which would have two different offxml files:
Jessica's experiments on per-molecule torsions: These would have an OFFXML file which only adds TIG0 and go one molecule at a time (and your TIG0 would not be interpolated, since interpolation is only important across multiple molecules)
A tool that
Takes as input a QM TorsionDrive, and runs forcebalance with Parsley + TIG0, where forcebalance optimizes the k values for TIG0
Pavan’s experiments, which fit many molecules at a time and have a full range of TIGs
Targets: In Jessica’s case she would fit to an individual torsion drive for an individual molecule. Then she would iterate through this, iterating over molecules/torsion drives — for each, getting a new TIG0 force constant for each molecule.
In Pavan’s case he’d be fitting to many torsion drives (potentially all the data he’s been plotting, except those molecules used in fitting) simultaneously."
1 5
\ /
2 – 4
/
3
Jeff’s idea
If we want to generate “barrier heights attributable to torsions”, then there are two ways to do it:
Use forcebalance to fit all the torsions about a particular bond using a QM torsiondrive, then look at the energy vs. angle profile that those refit torsions make, and find the difference between the highest and lowest points
Take the same QM torsiondrive, and calculate an MM energy for all QM conformations, but where the MM torsions about the central bond all have a magnitude of 0. Then make a plot of the QM - MM energy difference, and find the energy difference between the highest and lowest points.
I am in favor of the second idea. It has all the merits of the first idea, since this algorithm is effectively what ForceBalance would do internally when fitting the torsion parameters. It has the added benefit of not trying to show us the QM-minus-MM energy profile through the hazy filter of “trying to reproduce the plot using a bunch of cosines”, but instead it will show us the raw difference, which should be more suitable for finding the barrier height.
Trevor’s idea
The QM barrier height is a total height, which includes many effects, which may add unwanted noise and make linear correlations hard to establish. Using a current FF, we can estimate the exact torsion barrier height by calculating the dihedral terms. This of course ignores vdW which is generally highly coupled.
(After much discussion, many people presented big-picture ideas with varying amounts of overlap about how data analysis, infrastructure, and torsion interpolation could work, both on a conceptual and an technical level)
JM – Let’s make sure we use this time to identify current infrastructure blockers and how to resolve them. I have the tooling that I need currently.
PB – With some assistance from Trevor, I also have the tooling I need currently.
JW – Great. Let’s stay in touch as the needs/directions for this project evolve so we can ensure it doesn’t get too hung up by infrastructure.
Discussion topics
Time | Item | Presenter | Notes |
|---|---|---|---|
|
|
| |