RNA has been highlighted as a promising therapeutic target extending beyond the traditional protein-centric paradigm of drug development. 1 For RNA-targeting small-molecule drug design, an accurate prediction of ligand binding sites is essential, serving as a strategic starting point for structure-based approaches. RNA secondary structures, such as loops, are closely associated with functional binding regions, and due to RNA’s asymmetric three-dimensional structure with chiral centers, geometry-aware learning provides more effective modeling of the spatial features for defining binding pockets. Recent advances in structural analysis tools, such as FreeSASA and Ghecom, have enabled systematic characterization of RNA surface areas and cavity features. DSSR also provides detailed annotations of RNA secondary structures, as well as interaction patterns such as stacking and chemical contacts, which are critical for understanding ligand binding. In this study, we propose a R-GAMBIT, that predicts RNA-small molecule binding nucleotides using geometric tensor network which achieves E(3)-equivariance. The model leverages RNA-specific geometric and topological features to identify binding pockets more effectively. Comprehensive benchmark evaluations on TE18, RB9, JL10, and TL12 data sets demonstrate that our method consistently outperforms existing approaches, achieving state-of-the-art performance. These results highlight the model’s potential as a robust technical foundation for precise target identification and structure-based drug design in RNA-targeted therapeutics.
JinHyeok Yoo (Sun,) studied this question.