Significant Subgraph Detection in Multi-omics Networks for Disease Pathway Identification.

Area of Science:

• Biomedical Informatics
• Network Biology
• Computational Biology

Background:

• Chronic obstructive pulmonary disease (COPD) is a leading cause of death, necessitating research into complex disease pathways.
• Identifying networks of interacting biomarkers is crucial for understanding COPD progression and discovering treatments.
• Sparse multiple canonical correlation network analysis (SmCCNet) models relationships between omics data and disease phenotypes.

Purpose of the Study:

• To introduce novel methods for detecting significant subgraphs in multi-omics networks.
• To overcome limitations of hierarchical clustering in identifying biologically relevant subgraphs.
• To improve the quality and comprehensiveness of detected disease-associated biological networks.

Main Methods:

• Developed two new subgraph detection algorithms: Correlated PageRank and Correlated Louvain.
• Extended existing Personalized PageRank Clustering and Louvain algorithms for network analysis.
• Compared the performance of the new methods against the current hierarchical clustering approach.

Main Results:

• The proposed Correlated PageRank and Correlated Louvain methods significantly improved subgraph quality.
• These novel techniques offer better identification of complex biological relationships compared to hierarchical clustering.
• The hybrid approach combining both methods also demonstrated superior performance.

Conclusions:

• Correlated PageRank and Correlated Louvain are effective alternatives to hierarchical clustering for significant subgraph detection.
• These advanced methods enhance the exploration of multi-omics data for diseases like COPD.
• Improved subgraph detection facilitates a deeper understanding of disease mechanisms and biomarker discovery.