An Integrated TCN-CrossMHA Model for Predicting circRNA-RBP Binding Sites

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Circular RNAs (circRNAs) interact with RNA-binding proteins (RBPs), influencing disease development.
• Accurate prediction of circRNA-RBP binding sites is crucial for understanding disease mechanisms and developing therapeutic strategies.

Purpose of the Study:

• To develop a novel computational approach, circTCA, for predicting circRNA-RBP binding sites.
• To enhance the understanding of circRNA-RBP interactions for potential disease treatment applications.

Main Methods:

• Utilized temporal convolutional networks (TCN) and a cross multi-head attention mechanism.
• Employed two distinct encoding strategies for circRNA sequences.
• Implemented global expectation pooling and a fully connected layer for classification.

Main Results:

• The proposed circTCA method demonstrated effectiveness in predicting circRNA-RBP binding sites.
• Comparative analysis against five other methods and ablation experiments confirmed circTCA's superior performance.
• Feature visualization and motif analysis validated the model's predictive capabilities.

Conclusions:

• circTCA is an effective tool for predicting circRNA-RBP binding sites.
• The developed method offers valuable insights into molecular interactions relevant to disease pathogenesis.
• This approach can contribute to the development of novel therapeutic strategies targeting circRNA-RBP interactions.