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

The article focuses on the development and validation of NICE-FF (Non-empirical, Intermolecular, Consistent and Extensible Force Field), a new ab initio intermolecular force field parametrized via machine learning (ML) techniques for nucleic acids and related organic molecules. NICE-FF was trained on extensive dimer interaction energy datasets computed at the spin component scaled-MI-second order Møller–Plesset perturbation theory (SCS-MI-MP2) level and employs a buffered 14-7 potential form. Its performance was benchmarked against popular classical force fields using the S22 dataset, where NICE-FF demonstrated superior accuracy in reproducing high-level ab initio interaction energies. The force field was further validated through successful predictions of two- and three-dimensional crystal structures of DNA bases and other organic molecules, including hypoxanthine, uracil, pyrazinamide, 9-methylhypoxanthine, and theophylline, using the in-house crystal structure prediction tool FFCASP. The study highlights NICE-FF's extensibility to new atom types and its potential for broader application in modeling organic molecular crystals with improved accuracy and computational efficiency.