Predicting circRNA-miRNA interactions utilizing transformer-based RNA sequential learning and high-order proximity...

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• Circular RNAs (circRNAs) regulate gene expression by sponging microRNAs (miRNAs), making their interactions critical for medical research.
• Current challenges include a scarcity of verified circRNA-miRNA interactions for model training and a lack of interpretability in existing prediction models.

Purpose of the Study:

• To develop an interpretable and high-performing method for predicting circRNA-miRNA interactions.
• To address the limitations of small training datasets and enhance model transparency in circRNA-miRNA interaction prediction.

Main Methods:

• Proposed SPBCMI, a method integrating sequence features from Bidirectional Encoder Representations from Transformer (BERT) and structural features from graph embedding.
• Utilized a Gradient-boosted decision trees (GBDT) classifier for predicting circRNA-miRNA interactions.
• Evaluated model performance using Area Under the Curve (AUC) and a case study on predicting known interactions.

Main Results:

• Achieved a state-of-the-art AUC of 0.9143 for circRNA-miRNA interaction prediction.
• SPBCMI accurately predicted 7 out of 10 circRNA-miRNA interactions in a case study.
• Demonstrated superior performance compared to existing methods in predicting these interactions.

Conclusions:

• SPBCMI offers an innovative and high-performing solution for predicting circRNA-miRNA interactions.
• The method enhances understanding of circRNA-miRNA regulatory mechanisms.
• SPBCMI provides a valuable tool for medical research involving circRNA and miRNA.