Accurate and transferable classical force fields for noncovalent interactions

Establish accurate prediction of noncovalent interactions across broad regions of chemical space using classical force fields, addressing the persistent challenge of achieving reliable accuracy for applications such as protein–ligand binding and crystal structure prediction.

Background

The paper motivates DensIP by noting that accurate modeling of noncovalent interactions is important in protein–ligand binding, crystal structure prediction, and related applications. Although machine-learned force fields can achieve high accuracy when supplied with sufficiently large training datasets, generating high-quality coupled-cluster reference data remains computationally expensive.

The authors identify the development of accurate and transferable classical force fields as an unresolved challenge. DensIP is presented as a physics-based, density-driven approach intended to address this broader problem while reducing the amount of high-level reference data required.