Unsupervised discovery of phenotype-specific multi-omics networks.

Area of Science:

• Genomics and Systems Biology
• Computational Biology
• Network Medicine

Background:

• Complex diseases manifest diverse phenotypic traits, necessitating integrated biological understanding.
• Current omics data integration methods often overlook quantitative phenotypic trait incorporation.
• Understanding trait-specific biological mechanisms is crucial for disease etiology and targeted therapies.

Purpose of the Study:

• To introduce sparse multiple canonical correlation network analysis (SmCCNet), a novel technique for multi-omics data integration with quantitative phenotypes.
• To construct phenotype-specific multi-omics networks, focusing initially on miRNA-mRNA interactions.
• To evaluate SmCCNet's performance against existing network construction approaches.

Main Methods:

• Developed SmCCNet, a sparse multiple canonical correlation network analysis method.
• Applied SmCCNet to miRNA-mRNA network construction for chronic obstructive pulmonary disease (COPD) and breast cancer.
• Validated SmCCNet using simulations and real-world disease data.

Main Results:

• SmCCNet demonstrated superior prediction performance over popular gene expression network methods in simulations.
• Analysis of COPD revealed enrichment in the Cadherin pathway.
• Breast cancer analysis highlighted the interferon-gamma signaling pathway and novel omics features.

Conclusions:

• SmCCNet effectively integrates multi-omics data with phenotypic traits for phenotype-specific network construction.
• The method identified known and novel biologically relevant pathways in COPD and breast cancer.
• SmCCNet is versatile, applicable to various data types, and extensible for multiple phenotypes.