JLCRB: A unified multi-view-based joint representation learning for CircRNA binding sites prediction

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Circular RNAs (CircRNAs) play a key role in gene regulation by binding to RNA Binding Proteins (RBPs).
• Accurate identification of RBP binding sites on CircRNAs is essential for understanding disease mechanisms.
• Current multi-view methods for RBP-CircRNA binding site prediction face challenges with increasing feature dimensions and ignoring intrinsic data connections.

Purpose of the Study:

• To develop a novel multi-view joint representation learning network for enhanced RBP binding site prediction on CircRNAs.
• To address limitations of existing methods by improving the consistency and integration of multi-view feature information.
• To provide a robust computational tool for RBP-CircRNA interaction analysis.

Main Methods:

• Proposed a multi-view joint representation learning network utilizing diverse feature encoding methods for CircRNA data.
• Constructed intrinsic connections between views by generating a global joint representation for feature calibration.
• Employed fused multi-view depth features for accurate detection of RBP binding sites on CircRNAs.

Main Results:

• Achieved an average AUC of 93.68% across 37 CircRNA-RBP datasets.
• Demonstrated superior prediction performance compared to existing state-of-the-art methods.
• The developed method effectively highlights important features and suppresses irrelevant ones through feature calibration.

Conclusions:

• The proposed multi-view joint representation learning network significantly improves the prediction of RBP binding sites on CircRNAs.
• This approach offers a more effective way to integrate and leverage multi-view data for biological sequence analysis.
• The study provides valuable computational resources, including code and a web server, for the research community.