Graph Neural Networks for Structure-Aware RNA-RNA Interaction Modeling

RNA-RNA interactions (RRIs) are inherently structural phenomena, governed not only by linear sequence complementarity but also by secondary and tertiary folding patterns that bring distant nucleotides into spatial proximity. While sequence-based deep learning models have demonstrated promising performance, they often struggle to explicitly encode RNA structural information. Graph Neural Networks (GNNs) provide a natural and flexible framework for modeling RNA molecules as graphs, where nucleotides are represented as nodes and structural or spatial relationships are encoded as edges. This chapter presents a comprehensive treatment of GNN-based approaches for structure-aware RRI prediction. We discuss graph representations of RNA sequences and structures, graph construction strategies from predicted and experimentally derived structures, and state-of-the-art GNN architectures including Graph Convolutional Networks, Graph Attention Networks, and Graph Transformers. Case studies and benchmarking results demonstrate the advantages of graph-based models in capturing long-range and non-local RNA interaction patterns. Challenges and future directions for scalable and interpretable graph-based RRI modeling are also discussed.