MOTGNN: Interpretable Graph Neural Networks for Multi-Omics Disease Classification.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Multi-omics data integration (DNA methylation, mRNA, microRNA) offers insights into disease mechanisms.
• Challenges include high dimensionality, data heterogeneity, and lack of reliable networks.
• Existing models often lack interpretability and are vulnerable to class imbalance.

Purpose of the Study:

• To propose MOTGNN, a novel and interpretable framework for binary disease classification using multi-omics data.
• To address limitations of existing methods, including handcrafted graphs and class imbalance.

Main Methods:

• MOTGNN framework utilizes eXtreme Gradient Boosting (XGBoost) for supervised graph construction.
• Employs modality-specific Graph Neural Networks (GNNs) for hierarchical representation learning.
• Integrates omics data using a deep feedforward network.

Main Results:

• MOTGNN achieved 5-10% higher accuracy, ROC-AUC, and F1-score across three disease datasets compared to state-of-the-art baselines.
• Demonstrated robustness to severe class imbalance.
• Provided built-in interpretability, identifying key biomarkers and modality contributions.

Conclusions:

• MOTGNN offers a significant advancement in multi-omics data integration for disease classification.
• The framework enhances predictive accuracy while maintaining computational efficiency and interpretability.
• MOTGNN holds potential for improving biomedical applications through better disease modeling.