/
SUMMER CAMP PROJECT: TorsionDrive minimal basis set with existing bonds and angles

SUMMER CAMP PROJECT: TorsionDrive minimal basis set with existing bonds and angles

 

Driver

Approver

Contributors

 

Other stakeholders

@Michael Shirts @Julianne Hoeflich

Objective

Experiment with fitting torsions initially to a “minimal basis set” of torsiondrives, before a fuller fit

Time frame

 

Key outcomes

  • a force field fit with smee to a final torsiondrive dataset

  • a force field initially fit to a minimal basis set of torsions

  • a force field initially fit to a minimal basis set of torsions, then to a final torsiondrive dataset

Key metrics

  • performance on training set of TorsionDrives (YAMMBS)

  • performance on an expansive set of TorsionDrives (YAMMBS)

  • performance on particular benchmarks (simple alkanes)

Status

Not started In progress Stalled Complete

GitHub repo

Slack channel

Designated meeting

Released force field

Publication

 Problem Statement and Objective

In ForceBalance training, we have observed torsion profiles that differ substantially from QM despite training. One hypothesis is that this is due to torsions getting stuck in local minima; in ForceBalance, we allow torsion parameters to drift from force field to force field instead of re-setting them through methods such as the modified Seminario method (MSM). One way of getting out of this local minimum may be to initially over-fit torsions to simple small molecules with low intra-torsion interactions, before re-fitting to the final larger dataset.

Objective: compare the performance of fitting initially to a “minimal basis set” of torsiondrive structures, compared to just fitting directly to the final dataset.

Note: this is a summer camp project. In addition, this project will be carried out using smee, where we have not validated performance on training to targets such as TorsionDrive profiles, or derived a dataset at the OpenFF level of theory that yields good performance in an existing fit. As such the outcome of this project may be inconclusive.

 Scope

Must have:

  • A minimal basis set of structures for each torsion

  • A more expansive set of structures for each torsion

  • A force field fit to the minimal basis set

  • A force field fit to the larger dataset

  • A force field fit first to the minimal basis set, then to the larger dataset

  • all of the above, for a suitable problem space (e.g. hydrocarbon compounds)

Must have:

  • A minimal basis set of structures for each torsion

  • A more expansive set of structures for each torsion

  • A force field fit to the minimal basis set

  • A force field fit to the larger dataset

  • A force field fit first to the minimal basis set, then to the larger dataset

  • all of the above, for a suitable problem space (e.g. hydrocarbon compounds)

Nice to have:

  • all of the above, for the entire chemical space of the FF

Not in scope:

  • ForceBalance

Project Approaches

References