GONNMDA: A Ordered Message Passing GNN Approach for miRNA-Disease Association Prediction.

Area of Science:

• Biochemistry
• Genomics
• Computational Biology

Background:

• MicroRNAs (miRNAs) are crucial non-coding molecules in disease diagnosis and prognosis.
• Traditional wet-lab validation of miRNA-disease associations is inefficient.
• Deep learning offers advanced tools for uncovering miRNA-disease patterns, but existing methods struggle with multi-molecular interactions and graph complexities.

Purpose of the Study:

• To develop an advanced deep learning model for accurate prediction of miRNA-disease associations.
• To overcome limitations of existing methods, including handling heterogeneous data and the over-smoothing problem in graph neural networks.
• To provide a reliable computational tool for identifying novel miRNA-disease relationships.

Main Methods:

• The GONNMDA model integrates multi-source similarity features and applies noise reduction for a comprehensive representation.
• It constructs heterogeneous graphs and employs root-tree hierarchical alignment with ordered gating message passing.
• A multilayer perceptron is utilized for final association predictions.

Main Results:

• GONNMDA achieved high predictive performance with an Area Under the Curve (AUC) of 95.49% and an Area Under the Precision-Recall Curve (AUPR) of 95.32%.
• The model demonstrated superior performance compared to several state-of-the-art methods.
• Case studies and survival analyses on breast, rectal, and lung cancers validated GONNMDA's effectiveness and reliability.

Conclusions:

• GONNMDA effectively predicts miRNA-disease associations by addressing data heterogeneity and over-smoothing challenges.
• The model offers a significant advancement over existing computational approaches for miRNA-disease association prediction.
• GONNMDA shows promise for accelerating biomarker discovery and understanding disease mechanisms in oncology.