CircRNA identification and feature interpretability analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNAs (circRNAs) play roles in microRNA regulation and disease, including cancer.
• Accurate identification of circRNAs is crucial for functional research but current prediction models lack sufficient accuracy.
• A need exists for improved computational frameworks for circRNA prediction and analysis.

Purpose of the Study:

• To develop a novel and accurate computational framework for classifying circular RNAs (circRNAs) from other long non-coding RNAs (lncRNAs).
• To enhance the understanding of circRNA functions and their involvement in biological processes and diseases.

Main Methods:

• Developed CircDC, a novel framework utilizing four feature encoding schemes.
• Employed a multilayer convolutional neural network and bidirectional long short-term memory network for high-order feature representation and prediction.
• Performed interpretable analysis of features influencing model performance.

Main Results:

• CircDC demonstrated superior accuracy in circRNA prediction compared to existing models.
• The framework successfully identified circRNAs and provided insights into feature importance.
• Applied the computational framework for extended circRNA identification.

Conclusions:

• CircDC is an effective tool for circRNA prediction, facilitating deeper understanding of related biological functions.
• Feature importance analysis enhances model interpretability and reveals significant biological properties.
• The developed code and data are publicly available for research use.