Development of deep learning methods for RBP binding sites prediction

Predicting the binding sites of RNA-binding protein (RBP) plays a crucial role in understanding how RBP participates in gene regulation. In recent years, with the accumulation of a large amount of high-throughput experimental data and the rapid development of deep learning, deep learning methods have been more and more widely used in the field of RBP binding sites prediction. Deep learning models can detect hidden patterns in massive biological data, and has the advantages of high speed and high robustness compared with traditional experimental methods. The experiments of various studies have proved that deep learning methods have achieved remarkable performance and they are gradually improving. In this review, commonly used RNA-protein binding site databases are summarized, and classic deep learning models including RNA sequence coding methods are introduced, and the successful applications of deep learning in the field of RBP binding site prediction in recent years are mainly reviewed. Then the RBP binding motif mining methods are further summarized. Finally, the limitations and potentials of the current deep learning methods in the application of RBP binding site prediction, as well as potential improvements are discussed.