Heterogeneous Graph Convolutional Networks and Matrix Completion for miRNA-Disease Association Prediction

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Biological experiments are costly and complex, driving the need for computational methods to predict microRNA-disease associations.
• Existing computational methods face challenges due to the intricate relationships between microRNAs and diseases, requiring consideration of both local and global network influences.
• Predicting disease-related microRNAs with no known associations is a critical unmet need.

Purpose of the Study:

• To develop a novel computational method for predicting microRNA-disease associations.
• To construct a heterogeneous network integrating miRNA similarity, disease similarity, and known associations.
• To leverage graph convolutional networks (GCNs) for enhanced information aggregation within the network.

Main Methods:

• Constructed a heterogeneous network combining miRNA similarity (including families and clusters), disease similarity, and known miRNA-disease associations.
• Utilized graph convolutional networks (GCNs) to aggregate neighborhood information for each node, capturing local and global influences.
• Evaluated the method using rigorous cross-validation techniques (5-fold, leave-one-disease-out, global/local leave-one-out).

Main Results:

• The proposed method achieved high performance with Area Under the Curve (AUC) values of 0.9616, 0.9946, 0.9656, and 0.9532 across different cross-validation schemes.
• Demonstrated significant outperformance compared to existing state-of-the-art methods.
• Case studies confirmed the method's efficacy in predicting novel disease-related microRNAs without prior associations.

Conclusions:

• The developed computational approach effectively predicts microRNA-disease associations by integrating diverse data sources within a heterogeneous network.
• Graph convolutional networks enhance the prediction accuracy by capturing complex network topology and information propagation.
• This method offers a powerful tool for identifying potential microRNA biomarkers for diseases, particularly for those lacking existing associations.