A Novel Deep Learning Method for Predicting RNA-Protein Binding Sites

The cell cycle and biological processes rely on RNA RNA-binding protein (RBP) interactions. It is crucial to identify the binding sites RNA. Various deep-learning methods have been used for site prediction. However, they cannot extract hierarchical features of secondary structure. Therefore, this paper proposes HPNet, which can automatically -binding preferences. HPNet performs feature learning from two perspectives sequence A convolutional neural network (CNN), a method, learn in HPNet. To capture information RNA, we introduced DiffPool into differentiable pooling graph (GNN). CNN were combined improve prediction accuracy by leveraging both Binding preferences be extracted based model outputs parameters. Overall, experimental results showed that achieved mean area under curve (AUC) 94.5% benchmark dataset, was more accurate than state-of-the-art methods. Moreover, these demonstrate structure play an essential role selecting sites.