Heterogeneous Types of miRNA-Disease Associations Stratified by Multi-Layer Network Embedding and Prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Abnormal microRNA (miRNA) functions are implicated in numerous diseases, with complex associations recorded in the Human miRNA-Disease Associations (HMDD) database.
• These associations can be heterogeneous, encompassing genetics, epigenetics, circulating miRNAs, tissue expression, and miRNA-target interactions.
• Predicting unknown or novel miRNA-disease associations is crucial for understanding disease mechanisms.

Purpose of the Study:

• To develop a novel method for predicting miRNA-disease association types.
• To leverage the non-linear characteristics of the miRNA-disease association network for improved prediction accuracy.
• To identify new miRNA-disease associations beyond existing databases.

Main Methods:

• Proposed an attributed multi-layer heterogeneous network embedding method.
• Learned latent representations of miRNAs and diseases from various association types.
• Predicted the existence of association types for all miRNA-disease pairs.

Main Results:

• The proposed method demonstrated superior prediction performance compared to two recent methods via 10-fold cross-validation on the HMDD v3.2 database.
• Achieved accurate predictions under various settings, outperforming existing approaches.
• Real predictions made beyond the HMDD database were validated by NCBI literature.

Conclusions:

• The attributed multi-layer heterogeneous network embedding method effectively predicts miRNA-disease associations and their types.
• The approach accurately identifies novel associations, contributing to a deeper understanding of disease pathogenesis.
• This method offers a powerful tool for exploring the complex landscape of miRNA-disease relationships.