HeMDAP: Heterogeneous Graph Self-Supervised Learning for MiRNA-Disease Association Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MiRNA-disease associations (MDAs) are crucial for understanding human disease pathology.
• Traditional experimental methods for MDA identification are inefficient and costly.
• Existing machine learning methods for MDA prediction face limitations due to supervised learning constraints.

Purpose of the Study:

• To propose a novel method, HeMDAP, for accurate prediction of miRNA-disease associations.
• To overcome the limitations of supervised learning in current MDA prediction models.
• To leverage graph contrastive learning for improved prediction performance.

Main Methods:

• Developed HeMDAP, a method based on graph contrastive learning utilizing meta-path and network structure views of a heterogeneous graph.
• Employed self-supervised and supervised contrastive learning to optimize node embeddings.
• Integrated knowledge-aware enhancement to improve embedding quality.
• Utilized a multi-view learning and multi-task training strategy.

Main Results:

• HeMDAP demonstrated superior prediction accuracy compared to all existing methods on public datasets.
• Achieved an Area Under the Curve (AUC) of 94.92% in five-fold cross-validation.
• Achieved an Area Under the Precision-Recall Curve (AUPR) of 95.07% in five-fold cross-validation.

Conclusions:

• HeMDAP effectively captures complex relationships among miRNAs, genes, and diseases.
• The proposed multi-view learning and contrastive learning strategy significantly enhances MDA prediction.
• HeMDAP represents a superior approach for miRNA-disease association prediction.