Quantifying binding stability by clustering conformations to enhance binding prediction accuracy

Abstract

Accurate prediction of ligand–protein binding remains a central challenge in drug design. Incorporating bound-state dynamics promises enhanced accuracy, yet efforts have been hampered by an unclear relationship between binding stability and affinity and by the lack of efficient techniques to quantify these dynamics. Here we introduce ShakeIt, a portable, high-throughput method that quantifies binding stability by clustering conformations from molecular-dynamics simulations. Application of ShakeIt to the PDBbind-2020 dataset reveals that effective ligand–protein recognition depends on dynamic conformational matching, in which the ligand and protein undergo coordinated structural adjustments to achieve mutual fit. Incorporating ShakeIt stability scores into empirical docking and physics-based free-energy calculations reliably distinguishes true ligands from decoys and reduces false positives. Prospective screening of an 18 million-compound library against the GluN1/GluN3A NMDA receptor yielded two novel micromolar antagonists, providing valuable leads for this understudied target. ShakeIt thus offers a broadly applicable route to leverage conformational dynamics for more accurate binding prediction and accelerated drug discovery.