GNN4DM: a graph neural network-based method to identify overlapping functional disease modules

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Identifying disease modules in molecular networks is crucial for understanding disease mechanisms and finding therapeutic targets.
• Traditional methods face challenges with network complexity, overlapping modules, and integrating diverse genomic data.
• There is a need for automated approaches that effectively leverage extensive biological knowledge and data.

Purpose of the Study:

• To propose GNN4DM, a novel graph neural network-based model for automated discovery of overlapping functional disease modules.
• To integrate network topology with genomic data for enhanced gene representation and pathway alignment.
• To improve the interpretability and accuracy of disease module identification.

Main Methods:

• Developed GNN4DM, a graph neural network model integrating network topology and genomic data.
• Trained and evaluated the model using the DREAM benchmark and extended datasets (GWAS Atlas, FinnGen, DisGeNET).
• Assessed performance against state-of-the-art methods for disease module detection.

Main Results:

• GNN4DM outperforms several state-of-the-art methods in detecting biologically meaningful disease modules.
• The model successfully learns gene representations aligned with biological pathways, enhancing interpretability.
• Discovered two novel multimorbidity modules enriched across diverse, seemingly unrelated diseases.

Conclusions:

• GNN4DM provides an effective, automated approach for identifying overlapping functional disease modules.
• The method enhances biological interpretability by integrating network and genomic data.
• GNN4DM demonstrates significant potential for uncovering complex disease relationships and therapeutic targets.