RNA-低分子結合部位予測における計算的進歩

RNA-small molecule interactions are fundamental to cellular regulation and have emerged as highly attractive therapeutic targets. Despite their potential, discovering RNA-binding small molecules remains challenging due to RNA's intrinsic structural flexibility, transient and context-dependent binding pockets, and the limited availability of high-resolution complex structures. Computational prediction approaches have evolved from early statistical models relying on handcrafted descriptors to advanced machine and deep learning frameworks that integrate sequence, structural, energetic, and topological information. More recently, large language models have enabled the capture of long-range sequence dependencies and contextual patterns, complementing structure-based encoders for multimodal modeling of RNA- ligand interactions. In this review, we summarize the principles and current state of computational strategies for RNA-ligand binding site prediction, highlighting methodological evolution, multimodal feature integration, and persisting challenges, and we discuss emerging directions toward accurate, generalizable, and interpretable predictions to accelerate rational RNA-targeted drug discovery.