Predicting protein-RNA binding signals with a hybrid deep neural network

Abstract RNA-binding proteins play critical roles in post-transcriptional regulation by interacting with RNA to regulate various cellular processes. Inferring the binding signals and sequence patterns of these interactions is essential for elucidating the mechanisms underlying gene expression regulation. In this study, we present RBPSignal, a deep learning-based computational tool designed to predict RBP binding signals on RNAs and identify potential sequence motifs associated with these interactions. RBPSignal leverages the deep learning framework and trains on comprehensive eCLIP datasets, demonstrating an enhanced predictive accuracy. Furthermore, the integration of model interpretability through Integrated Gradients enables the detailed analysis of binding motif syntax. We validate the efficacy of RBPSignal on chromosome data and compare the discovered motifs with existing motif database, showcasing its ability not only to predict binding signals but also to uncover sequence patterns correlated with binding signals. Our findings provide insights into the sequence specificity of RBPs and explore the protein-protein interaction networks. RBPSignal serves as a valuable tool for exploring RBP-RNA interaction landscapes, facilitating further investigations into the regulatory networks underlying gene expression. The web server is freely available at http://www.csbio.sjtu.edu.cn/bioinf/RBPSignal/ .