Pseudoknots: Based RRI Detection Using Machine Learning Models

RNA-RNA interactions (RRIs) are fundamental to diverse cellular processes, including gene regulation, RNA splicing, editing, and translation. These interactions, whether intra- or inter-molecular, contribute to RNA functionality and cellular regulatory networks. A major structural component influencing RRIs is the presence of pseudoknots-non-nested base-pairing interactions that are critical to the function of ribozymes, viral RNAs, telomerase, and ribosomal RNA. Despite their biological importance, pseudoknots are frequently excluded from computational models due to the increased complexity they introduce to RNA secondary structure prediction and interaction detection. Traditional dynamic programming algorithms are limited to pseudoknot-free models, resulting in many RNA-RNA interaction prediction tools simplifying or ignoring pseudoknots altogether. This oversimplification can compromise the accuracy of predictions, particularly in scenarios where pseudoknotted structures are essential for biological function. Recent advancements in machine learning (ML) and deep learning offer new potential to model such complex RNA features. By leveraging large-scale sequence and structural data, ML models can implicitly learn interaction patterns associated with pseudoknotted regions, offering improved predictive power over thermodynamics-based methods alone. This chapter explores how pseudoknot-aware modeling can be integrated into ML-based RRI detection pipelines. We discuss the biological relevance of pseudoknots, computational challenges in modeling them, and the strategies for designing and extracting features that capture pseudoknot structures. We also survey current ML approaches, including classical and deep learning models, and examine how they incorporate structural information. Case studies are presented to demonstrate practical applications, followed by a review of benchmarking methods, datasets, and performance evaluation strategies. By combining biologically informed structure prediction with advanced ML techniques, this chapter outlines a framework for improving the fidelity of RRI detection in the presence of pseudoknots.