Molecular maps of diseases from omics data and network embeddings

Area of Science:

• Systems biology
• Bioinformatics
• Genomics and Proteomics

Background:

• Identifying disease-specific proteins and pathways is crucial for understanding disease mechanisms and developing therapeutics.
• Omics analyses often examine genes/proteins individually, limiting a systems-level biological perspective.
• Integrating omics data with protein-protein interaction networks can provide a more holistic view.

Purpose of the Study:

• To develop a method for integrating omics data with protein functional association networks.
• To construct comprehensive "disease maps" for various complex diseases.
• To improve the identification of disease-relevant proteins and pathways compared to omics analysis alone.

Main Methods:

• Integrated disease-specific omics data (genetics, transcriptomics, somatic mutation, proteomics) with the STRING functional association network.
• Utilized node2vec embedding to represent the protein network.
• Constructed disease maps for seven diseases across inflammatory, oncological, neurological, and vascular categories.
• Applied a linear model on network embeddings for protein identification and clustering for pathway analysis.

Main Results:

• The network embedding approach identified 2-4 times more known disease-relevant proteins than omics analysis alone, at equivalent specificity.
• Clustering of disease maps revealed shared functional modules (e.g., inflammation, cancer hallmarks) and disease-specific modules (e.g., keratinization in atopic dermatitis, ECM remodeling in aortic aneurysm).

Conclusions:

• Protein network embedding is a valuable strategy for analyzing omics data to deepen our understanding of disease mechanisms.
• This approach enhances the discovery of both shared and unique biological pathways implicated in diverse diseases.
• The constructed disease maps offer a systems-level view for advancing therapeutic development.