GGCRB: A Graph Neural Network Approach for Predicting CircRNA-RBP Interactions Using Structural and Sequence...

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Circular RNAs (circRNAs) and RNA-binding proteins (RBPs) are vital for gene regulation.
• Experimental identification of circRNA-RBP interactions is resource-intensive and inefficient.
• Existing computational methods often neglect circRNA structural characteristics, limiting prediction accuracy.

Purpose of the Study:

• To develop an advanced computational framework, GGCRB, for predicting circRNA-RBP binding sites.
• To integrate both sequence and structural information of circRNAs for improved prediction.
• To overcome the limitations of current methods by incorporating structural features.

Main Methods:

• GGCRB employs multiple sequence encoding schemes (HFN, ND, NCP, DPCP, Doc2Vec) and convolutional layers.
• Structural features are extracted using RNAstructure and modeled via graph convolutional and attention networks.
• Bidirectional LSTM and multihead attention modules capture global interactions, followed by pooling and softmax for prediction.

Main Results:

• GGCRB demonstrated superior performance compared to existing models across 16 benchmark datasets.
• Ablation studies confirmed the contribution of both sequence and structural features.
• Motif analyses validated the biological relevance of the predicted interactions.

Conclusions:

• Integrating sequence and structural information is crucial for accurate circRNA-RBP interaction prediction.
• GGCRB offers a powerful and effective computational tool for studying circRNA-RBP binding.
• The framework advances our understanding of gene regulation mediated by circRNA-RBP complexes.