Identifying Circular RNA and Predicting Its Regulatory Interactions by Machine Learning.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Circular RNAs (circRNAs) are long non-coding RNAs with crucial regulatory roles.
• Accurate identification and understanding of circRNA regulatory mechanisms are vital.
• Current computational tools for circRNA prediction require improvement in accuracy.

Purpose of the Study:

• To develop advanced machine learning frameworks for circRNA identification and regulatory information prediction.
• To enhance the accuracy of circRNA discrimination from other long non-coding RNAs (lncRNAs).
• To systematically predict circRNA interactions with microRNAs, RNA binding proteins, and transcriptional regulation.

Main Methods:

• Introduced circLGB, a LightGBM-based framework integrating sequence features, A-to-I deamination, A-to-I density, and IRES for circRNA identification.
• Developed circMRT, an ensemble machine learning framework utilizing sequence, graph, genome context, and regulatory features.
• Employed feature selection and ensemble methods for robust model performance.

Main Results:

• circLGB demonstrated superior performance in discriminating circRNAs from other lncRNAs.
• circMRT effectively predicted diverse regulatory information of circRNAs.
• Both developed algorithms outperformed existing state-of-the-art methods on benchmark datasets.

Conclusions:

• circLGB and circMRT provide powerful computational tools for circRNA research.
• These frameworks advance the understanding of circRNA biogenesis and regulatory networks.
• The developed tools are publicly available for the scientific community.