Energetics of Noncovalent Interactions of Protein-Ligand Complexes for Drug Discovery

Accurate modeling of noncovalent protein-ligand interactions is critical for applications such as enzyme engineering and drug discovery. Here, we present a data set of 14,905 protein-ligand interaction energies using experimental structures derived from HiQBind, a high-quality protein-ligand structural database, and subsequently fragmented into dimer configurations that are classified into noncovalent interaction (NCI) types, including hydrogen bonds, hydrophobic contacts, halogen bonds, salt bridges, cation-π, and π-π interactions. Each NCI category was further evaluated with energy decomposition analysis (EDA)¹ as a powerful framework to partition total protein-ligand energies into physically meaningful NCI components for electrostatics, Pauli repulsion, dispersion, polarization, and charge transfer. We further use this data to benchmark the performance of current classical force fields and the current state-of-the-art machine-learned interaction potential (MLIP). Together, this data set provides a quantitative quantum mechanical survey of protein-ligand energetics, offering new insights into the molecular origins of protein-ligand NCIs to inform drug design while establishing benchmarks for next-generation force field and MLIP development.