A Scalable Molecular Force Field Parameterization Method Based on Density Functional Theory and Quantum-Level Machine...

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Drug Discovery

Background:

• Accurate molecular force fields (FF) are crucial for molecular modeling but challenging to generate, especially for novel druglike molecules.
• Current quantum mechanics (QM) methods for FF parameterization are highly accurate but computationally expensive and slow, limiting their application.
• Existing FF often lack accuracy for diverse molecular structures, hindering reliable predictions.

Purpose of the Study:

• To develop an automated and efficient method for molecular force field (FF) parameterization.
• To enable the use of both high-accuracy quantum mechanics (QM) and faster neural network potentials (NNPs) for FF generation.
• To improve the accuracy and speed of FF parameterization for applications in computational drug discovery.

Main Methods:

• Developed an automated FF parameterization workflow integrating density functional theory (DFT) and neural network potentials (NNPs).
• Utilized torchani-ANI-1x NNP to approximate QM energies for FF parameter generation.
• Implemented the method in HTMD and made it available via PlayMolecule for broad accessibility.

Main Results:

• The NNP-based FF parameterization achieved significantly faster computation times compared to DFT methods.
• The generated FF parameters demonstrated higher accuracy than standard force fields like GAFF2.
• The automated method successfully parameterized small molecules, showcasing its potential for broader applications.

Conclusions:

• The presented automated FF parameterization method offers a faster and more accurate alternative to traditional QM calculations.
• This approach is expected to be highly valuable for computational structure-based drug discovery (SBDD).
• The tool's availability in HTMD and PlayMolecule facilitates its adoption in molecular modeling research.