Escaping Atom Types in Force Fields Using Direct Chemical Perception.

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Drug Discovery

Background:

• Traditional molecular mechanics force fields rely on discrete atom types for parameter assignment.
• This indirect chemical perception approach creates challenges in extending force fields and leads to parameter redundancy.
• Existing methods struggle with complex molecular environments and require extensive parameterization.

Purpose of the Study:

• To introduce a novel direct chemical perception approach for assigning molecular mechanics force field parameters.
• To develop a new, simplified force field format, the SMIRKS Native Open Force Field (SMIRNOFF).
• To demonstrate the efficiency and accuracy of SMIRNOFF compared to traditional methods.

Main Methods:

• Direct chemical perception operates directly on the molecular graph, bypassing atom typing.
• Parameters are assigned using SMIRKS (SMiles ARbitrary Target Specification) chemical substructure queries.
• A minimalist force field, SMIRNOFF99Frosst, was constructed and validated.

Main Results:

• The SMIRNOFF format, with a concise parameter file (~300 lines), covers over 99.98% of a 5 million molecule test set.
• SMIRNOFF99Frosst demonstrates accuracy comparable to the General Amber Force Field for hydration free energies and organic liquid properties.
• This approach significantly reduces force field complexity and enhances generalizability.

Conclusions:

• Direct chemical perception via SMIRKS offers a powerful and generalizable method for force field parameterization.
• The SMIRNOFF format provides a simplified, accurate, and extensible framework for molecular mechanics.
• SMIRNOFF99Frosst serves as a robust starting point for further force field development and applications.