Machine Learning Potentials for Biomaterials: Applications and Perspectives

The accurate computational modeling of biomaterials has long been constrained by the fundamental trade-off between the high efficiency of classical force fields and the high accuracy of first-principles methods. In recent years, the rapid development of machine learning potentials (MLPs) has provided a transformative pathway to bridge this gap, enabling atomistic simulations achieve accuracy close to density functional theory (DFT) while significantly reducing computational cost. In this review, we systematically summarize the fundamental concepts and theoretical foundations of MLPs, representative methodological frameworks, their applications in biomaterials, and future development directions. We first outline the physical basis and major methodological architectures of MLPs and discuss their relationships with conventional force fields and DFT-based approaches. We then review key applications of MLPs across diverse classes of biomaterials, including calcium phosphate and related bioceramics, silicate- and glass-based biomaterials, degradable metallic biomaterials, as well as metal-organic frameworks (MOFs) and related porous biomaterials. Finally, we discuss the key challenges and future perspectives of MLPs in biomaterials research and highlight their potential synergistic integration with experimental characterization and multiscale computational methods. This review aims to provide a systematic reference and forward-looking perspective for leveraging MLPs to accelerate the design and discovery of advanced biomaterials.