CR-deal: Explainable Neural Network for circRNA-RBP Binding Site Recognition and Interpretation

Area of Science:

• Computational Biology
• Genomics
• Molecular Biology

Background:

• Circular RNAs (circRNAs) are non-coding RNA molecules with a unique closed structure.
• Interactions between circRNAs and RNA-binding proteins (RBPs) are vital for biological functions and post-transcriptional regulation.
• Existing computational models for predicting circRNA-RBP interactions lack accuracy and interpretability.

Purpose of the Study:

• To develop an interpretable joint deep learning network, CR-deal, for predicting circRNA-RBP binding sites.
• To improve the accuracy of circRNA-RBP interaction prediction by integrating sequence and structural features.
• To provide insights into the functional mechanisms of circRNA-RBP interactions through interpretable predictions.

Main Methods:

• CR-deal employs a graph attention network to unify sequence and structural features.
• The model utilizes integrated gradient feature interpretation to infer marker genes and functional regions.
• Genome-wide circRNA binding event data was used for model training and validation.

Main Results:

• CR-deal demonstrated improved accuracy in predicting circRNA-RBP binding sites across 37 circRNA and 7 lncRNA datasets.
• The model successfully provided interpretable predictions, identifying key marker genes within binding sites.
• Functional structural regions involved in circRNA-RBP interactions were discovered using 5 circRNA datasets.

Conclusions:

• CR-deal offers a powerful and interpretable tool for predicting circRNA-RBP interactions.
• The findings enhance the understanding of circRNA functions, regulatory mechanisms, and their roles in diseases.
• CR-deal facilitates deeper biological insights into circRNA biology and disease pathogenesis.