Torsion Multiplicity Phase I

Owned by Brent Westbrook
Oct 21, 2024
1 min read

 

Driver

@Brent Westbrook

Approver

@Lily Wang @Alexandra McIsaac

Contributors

@Pavan Behara, Meghan Osato, Jessica Maat

Other stakeholders

@David Mobley @Michael Gilson @Michael Shirts @Daniel Cole

Objective

A force field with explicit multiplicities for all proper torsion parameters

Time frame

?

Key outcomes

  • A force field including:

    • Explicit multiplicities for all proper torsion parameters

  • Optimization and TorsionDrive datasets to cover the new parameters

Key metrics

  • Equivalent or improved performance in key metrics (DDE, RMSD, TFD) on the industry benchmarking dataset

Status

In progress

GitHub repo

https://github.com/ntBre/valence-fitting/

Slack channel

 

Designated meeting

FF Fitting (Wednesdays 1pm PT)

Released force field

Publication

 

 Problem Statement and Objective

Current Sage-line force fields have many SMIRKS patterns like [#1:1]-[#7X4,#7X3:2]-[#6X4;r3:3]-[*:4] (t59) that correspond to multiple multiplicities, which can be calculated as M = (Nc - 1) * (Nd - 1), where Nc is the number of bonds to atom C. The hypothesis is that enumerating these multiplicities explicitly (ie making two parameters for t59: [#1:1]-[#7X4:2]-[#6X4;r3:3]-[*:4] and [#1:1]-[#7X3:2]-[#6X4;r3:3]-[*:4]) will improve the quality of the force field.

 

image-20241021-160752.png
Biphenyl with its four torsions

 Scope

Must have:

  • Explicit multiplicities for all proper torsion parameters

  • New training data to cover the new parameters

  • Improved or equivalent force field performance compared to Sage 2.2.1

Must have:

  • Explicit multiplicities for all proper torsion parameters

  • New training data to cover the new parameters

  • Improved or equivalent force field performance compared to Sage 2.2.1

Nice to have:

  • Updated periodicities/torsion shapes

Not in scope:

  • Other force field changes

Project Approaches

References