/
TorsionProfile Restraints

TorsionProfile Restraints

It has been suggested that the use of restraints on the atoms not involved in the target torsion during Forcebalance optimization might affect the performance of the forcefield in recreating the QM PES outside of the fitting. To investigate this we have fit a collection of molecules from the BACE JACS benchmark system with and without the use of restraints. This is achieved by using the keyword restrain_k=0 on the torsion profile target. After the molecules have been refit a new torsiondrive was run using the same input conformers as used for the reference QM torsiondrive but with the new bespoke forcefield. The fitting was done using openff-bespokefit starting from openff-1.3.0.

The Forcebalance fits

Molecules with multiple rotatable bonds were fit simultaneously. Note that while the torsionprofile target in forcebalance calculates and reports the RMSD it is not used in the objective function to influence the parameters. Future investigations would need to implement this to see if it can help improve the performance of the fitting.

Here we show the results of the forcebalance fits comparing the RMSD with and without restraints.

BACE Molecule 0

Parent molecule

Fragments

 

State

Min(QM)

Min(MM)

Range(QM)

Range(MM)

Max-RMSD angstrom

Energy-RMSE kcal/mol

State

Min(QM)

Min(MM)

Range(QM)

Range(MM)

Max-RMSD angstrom

Energy-RMSE kcal/mol

initial (with restraints)

135

30

0 - 2.247

-0.031 - 2.360

0.022

0.0997

final (with restraints)

135

30

0 - 2.247

-0.196 - 2.854

0.022

0.3428

initial (without restraints)

135

-150

0 - 2.247

-0.197 - 2.854

0.022

0.3430

final (without restraints)

135

30

0 - 2.247

-0.032 - 2.360

0.022

0.0997

Restraints

Forcebalan