From static structures to dynamic landscapes: Generative artificial intelligence for protein conformational dynamics

Protein function emerges from ensembles of interconverting conformations, presenting challenges beyond static structure prediction. Although recent advances in generative artificial intelligence (AI) have transformed native-fold prediction, capturing conformational landscapes and binding-associated structural transitions remains a central limitation for mechanistic biology and structure-based drug discovery. Emerging generative models now aim to learn dynamic conformational distributions directly, enabling ensemble generation, ligand-responsive receptor sampling, and pathway-level inference. However, predictive fidelity is still constrained by limited physical grounding, incomplete kinetic realism, data imbalance, and uncertain calibration. This review summarizes key developments from 2023 to 2025, examines their methodological and practical limitations, and discusses how generative AI may evolve into a reliable framework for dynamic structural biology and mechanism-guided drug design.