Advancing Force Fields Parameterization: A Directed Graph Attention Networks Approach

Area of Science:

• Computational chemistry
• Molecular modeling
• Physical chemistry

Background:

• Force fields (FFs) are crucial for simulating molecular systems but require extensive parametrization.
• Current parametrization methods are empirical, time-consuming, and rely on heuristics and experimental/computational data.
• Automating FF parameter assignment is an active research area.

Purpose of the Study:

• To develop an automated, graph-based approach for deriving Generalized Amber Force Field (GAFF) parameters.
• To improve the accuracy and transferability of GAFF parameters.
• To investigate the influence of functional forms on FF parametrization.

Main Methods:

• A graph-based force field (GB-FFs) model was developed.
• Parameters were derived directly from chemical environments using directed molecular graphs.
• An end-to-end parametrization approach was implemented, aggregating information from molecular graphs.
• Simulation results were compared against original GAFF parametrization.

Main Results:

• The GB-FFs model demonstrated improved transferability of GAFF parameters.
• Enhanced accuracy was observed in modeling intermolecular and torsional interactions.
• Improved solvation free energies were achieved compared to standard GAFF.
• The method showed enhanced accuracy across a broader range of molecular complexes.

Conclusions:

• The proposed graph-based approach automates FF parameter derivation, reducing reliance on expert heuristics.
• The GB-FFs model significantly improves the accuracy and transferability of GAFF.
• This optimization approach is adaptable to other nonpolarizable and polarizable force fields.