PNAGMDA: A Principal Neighborhood Aggregation Based Graph Neural Network for miRNA-Disease Association Prediction.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial biomarkers for various diseases, with expression variations impacting disease pathways.
• Predicting miRNA-disease associations is vital for understanding disease mechanisms and developing targeted therapies.
• Existing graph neural network (GNN) models face limitations in capturing comprehensive node representations and diverse structural information.

Purpose of the Study:

• To develop an advanced computational method for accurate miRNA-disease association prediction.
• To overcome limitations of single GNN models and aggregation methods in learning node representations.
• To leverage integrated datasets and advanced network architectures for improved predictive performance.

Main Methods:

• Constructed a weighted heterogeneous graph integrating miRNA-LncRNA-disease interactions from multiple datasets.
• Employed Principal Neighborhood Aggregation (PNA) for robust node representation extraction using multiple aggregators.
• Fused features from PNA and Graph Attention Networks (GAT) via an attention mechanism for enhanced prediction.

Main Results:

• The proposed PNAGMDA method achieved high performance with Area Under the Curve (AUC) values of 93.82% on HMDD v2.0 and 92.77% on HMDD v3.2.
• Demonstrated superior accuracy in predicting miRNA-disease associations compared to existing methods.
• Case studies validated the reliability and effectiveness of PNAGMDA in identifying disease-associated miRNAs.

Conclusions:

• PNAGMDA offers a powerful and reliable approach for miRNA-disease association prediction.
• The integration of PNA and GAT effectively captures complex relationships within biological networks.
• This method holds significant potential for advancing biomarker discovery and personalized medicine.