HCLAMCMI: Prediction of circRNA-miRNA Interactions Based on Hypergraph Contrastive Learning and an Attention...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Circular RNA (circRNA)-microRNA interactions (CMIs) are crucial regulators of gene expression, cell proliferation, and tumorigenesis.
• Accurate CMI identification is vital for disease pathogenesis understanding and developing diagnostic/therapeutic strategies.
• Existing computational methods for CMI prediction have limitations in feature representation.

Purpose of the Study:

• To propose HCLAMCMI, an advanced computational model for predicting circRNA-miRNA interactions (CMIs).
• To improve the accuracy and efficiency of CMI identification compared to existing methods.
• To leverage hypergraph learning and attention mechanisms for enhanced feature representation.

Main Methods:

• Extracted features from adjacency, similarity, and heterogeneous networks of circRNAs, miRNAs, and diseases.
• Constructed complementary hypergraphs to capture high-order relational information.
• Employed hypergraph convolutional networks, contrastive learning, and a channel attention mechanism for feature generation and refinement.

Main Results:

• HCLAMCMI achieved Area Under the Curve (AUC) and Area Under the Precision-Recall Curve (AUPR) values exceeding 0.98 on training data.
• The model demonstrated strong performance on independent test sets with AUC and AUPR values around 0.97.
• HCLAMCMI consistently outperformed existing CMI prediction models.

Conclusions:

• The proposed HCLAMCMI model offers a significant advancement in computational CMI prediction.
• Integrating hypergraph-based learning with attention mechanisms enhances feature representation and prediction accuracy.
• HCLAMCMI provides a robust tool for CMI identification, supporting disease research and therapeutic strategies.