A machine learning framework for accurately recognizing circular RNAs for clinical decision-supporting.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Circular RNAs (circRNAs) are non-coding RNA molecules with a closed-loop structure, distinct from linear transcripts.
• circRNAs are increasingly implicated in complex diseases, necessitating accurate detection methods.
• Existing computational methods for circRNA detection often suffer from high false positive rates, hindering clinical applications.

Purpose of the Study:

• To address the limitations of existing circRNA detection tools, particularly their high false positive rates.
• To develop a robust computational approach for accurately distinguishing true circRNAs from false positives in RNA-seq data.
• To enhance the reliability of circRNA identification for clinical decision-support systems.

Main Methods:

• Reviewed existing circRNA detection strategies and identified limitations of single-feature approaches.
• Developed CIRCPlus2, a machine learning framework utilizing multiple data signals from RNA-seq data.
• Implemented CIRCPlus2 using a Gradient Boosting Decision Tree (GBDT) framework for efficient circRNA recognition.

Main Results:

• CIRCPlus2 effectively filters false positive circRNAs by integrating multiple data signals.
• Experimental validation on real and simulation datasets demonstrated significant improvements in specificity.
• The approach maintained high sensitivity while substantially reducing false positives.

Conclusions:

• Filtering false positives is a critical step in RNA-seq data analysis for accurate circRNA identification.
• Machine learning frameworks, such as CIRCPlus2, are well-suited for addressing the challenge of false positive reduction.
• CIRCPlus2 provides an efficient and accurate method for distinguishing true circRNAs from false positives, aiding clinical decision-making.