PMDAGS: Predicting miRNA-Disease Associations With Graph Nonlinear Diffusion Convolution Network and Similarities.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial regulators of biological processes and serve as potential noninvasive biomarkers for disease diagnosis and prognosis.
• Computational methods are vital for identifying miRNA-disease associations to improve disease management.

Purpose of the Study:

• To introduce PMDAGS, a novel computational method for predicting miRNA-disease associations.
• To leverage graph nonlinear diffusion convolution networks and similarity measures for enhanced prediction accuracy.

Main Methods:

• Calculated miRNA and disease similarity using miRNA-target interactions, gene-disease associations, and known miRNA-disease associations.
• Constructed initial node features by combining similarity vectors with known association vectors.
• Applied a nonlinear diffusion graph convolution network to extract feature embeddings, followed by a multi-layer perceptron for association prediction.

Main Results:

• PMDAGS achieved high Area Under the Curve (AUC) values, including 0.9222 (5-fold cross-validation) on HMDD v2.0 and 0.9366 (5-fold cross-validation) on HMDD v3.2.
• The method demonstrated superior performance compared to existing computational approaches in predicting miRNA-disease associations.
• Cross-validation results (5CV, 10CV, GLOOCV) consistently validated the effectiveness of PMDAGS on both HMDD v2.0 and HMDD v3.2 datasets.

Conclusions:

• PMDAGS effectively predicts potential miRNA-disease associations.
• The proposed method offers a significant advancement in computational approaches for biomarker discovery and disease association studies.
• PMDAGS outperforms existing methods, highlighting its potential for clinical applications in disease diagnosis and prognosis.