Hybrid Deep Learning for RNA-Protein Binding Prediction with Hierarchical Features and Debiasing

RNA-binding proteins (RBPs) are crucial for numerous cellular processes, with precise identification of RNA-protein binding sites being fundamental to understanding gene regulation mechanisms. This study introduces a novel hybrid model that leverages a hierarchical representation of RNA secondary structures and incorporates a context-average debiasing technique to enhance prediction accuracy. By synergizing Convolutional Neural Networks (CNN), Graph Neural Networks (GNN), and Bidirectional Long Short-Term Memory (BiLSTM) networks, This approach captures nuanced sequence and structural characteristics of RNA binding regions. Tested on the RBP-24 dataset, this model demonstrates superior performance, achieving significant improvements in prediction accuracy and computational efficiency compared to existing deep learning models.