Deep learning on graphs for multi-omics classification of COPD.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Network approaches are vital for understanding complex diseases like COPD.
• Integrating protein-protein interaction (PPI) networks with omics data for disease prediction remains a challenge.
• Deep learning, particularly Convolutional Graph Neural Networks (ConvGNNs), shows promise for disease classification using transcriptomics and PPI data.

Purpose of the Study:

• To develop a novel deep learning model for Chronic Obstructive Pulmonary Disease (COPD) classification by integrating multi-omics data and a reconstructed PPI network.
• To improve the accuracy of COPD prediction by leveraging advanced network reconstruction and deep learning techniques.
• To enhance the interpretability of deep learning models in disease prediction using explainable AI methods.

Main Methods:

• Reconstructed a COPD-specific PPI network using the AhGlasso algorithm on transcriptomics data.
• Extended ConvGNN methods to integrate PPI, proteomics, and transcriptomics data for COPD classification.
• Applied SHapley Additive exPlanations (SHAP) to interpret the ConvGNN model's predictions and identify key genes.

Main Results:

• The integrated deep learning model achieved higher accuracy in COPD classification compared to conventional methods.
• The AhGlasso-derived COPD-associated network further boosted prediction accuracy.
• SHAP analysis identified key COPD-related genes (e.g., CXCL11, IL-2, CD48) and enriched pathways (glycosaminoglycan, heparin signaling).

Conclusions:

• Integrating multi-omics data with reconstructed PPI networks via deep learning offers a powerful approach for COPD prediction.
• The developed model provides a more accurate and interpretable framework for understanding COPD mechanisms.
• This study highlights the potential of AI-driven network biology in advancing precision medicine for complex respiratory diseases.