NICE-FF: A non-empirical, intermolecular, consistent, and extensible force field for nucleic acids and beyond

Area of Science:

• Computational Chemistry
• Molecular Modeling
• Biophysics

Background:

• Accurate intermolecular force fields are crucial for molecular simulations of nucleic acids.
• Existing force fields often lack accuracy for complex systems and require extensive parametrization.
• Developing novel, efficient, and accurate force fields is an ongoing challenge in computational chemistry.

Purpose of the Study:

• To develop a new non-empirical ab initio intermolecular force field (NICE-FF) for nucleic acids and related molecules.
• To implement an automated framework for force field generation and parametrization.
• To validate the performance of NICE-FF against high-level ab initio calculations and experimental data.

Main Methods:

• Utilized spin component scaled-MI-second order Møller-Plesset perturbation theory for ab initio calculations.
• Employed machine learning for force field parametrization using a dataset of ~18,000 dimer geometries.
• Integrated NICE-FF into a crystal structure prediction (CSP) tool (FFCASP).

Main Results:

• Achieved a mean square deviation of ~0.46 kcal/mol during machine learning-assisted parametrization.
• NICE-FF demonstrated superior performance on the S22 dataset compared to established force fields.
• Successfully predicted experimental crystal structures for various molecules using the FFCASP tool.

Conclusions:

• NICE-FF provides highly consistent interaction energies with high-level ab initio results.
• The automated framework facilitates efficient force field development and extension.
• NICE-FF shows significant potential for accurate crystal structure prediction of nucleic acids and related compounds.