Predicting disease-associated circular RNAs using deep forests combined with positive-unlabeled learning methods.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Circular RNAs (circRNAs) are increasingly recognized for their roles in disease.
• Limited experimentally validated disease-associated circRNAs hinder computational method development.
• Systematic prediction approaches for disease-associated circRNAs are currently lacking.

Purpose of the Study:

• To develop and validate a novel computational framework for predicting disease-associated circRNAs.
• To leverage deep forests and positive-unlabeled learning for enhanced prediction accuracy.
• To provide a new tool for investigating circRNA-disease associations and functions.

Main Methods:

• Construction of a heterogeneous biological network integrating circRNAs, miRNAs, and diseases.
• Extraction of 24 meta-path-based topological features from the network.
• Application of deep forests combined with positive-unlabeled learning for prediction.
• Performance evaluation using 5-fold cross-validation on 15 disease datasets with Recall@k and PRAUC@k metrics.

Main Results:

• The proposed deep forest-based method demonstrated superior performance compared to other competitive methods.
• Performance of all evaluated methods improved with the inclusion of more known positive labels.
• The study successfully identified potential disease-related circRNAs through a systematic approach.

Conclusions:

• The developed framework offers a robust method for predicting disease-associated circRNAs.
• This approach can significantly improve the understanding of circRNA functions in disease pathogenesis.
• The findings pave the way for future research in circRNA-based diagnostics and therapeutics.