Molecular heterogeneity at the network level: high-dimensional testing, clustering and a TCGA case study

Area of Science:

• Computational Biology
• Network Medicine
• Statistical Genomics

Background:

• Molecular networks are crucial in biology and disease, but detecting differences between contexts (e.g., cancer types) is challenging.
• Existing methods primarily focus on expression levels, overlooking network structure variations.
• Identifying subgroups based on network differences is an unsupervised learning problem.

Purpose of the Study:

• To develop statistical methods for testing and clustering network differences in biological data.
• To identify molecular subgroups and their specific network characteristics.
• To provide a computational tool for network analysis in disease.

Main Methods:

• Leveraging high-dimensional statistics for network analysis.
• Applying methods to continuous molecular measurements without pre-defined networks.
• Utilizing the Bioconductor package nethet for implementation.

Main Results:

• Demonstrated significant differences in signaling protein interaction networks across cancer types using TCGA data.
• Showcased the ability of the proposed methods to identify cancer subtypes.
• Validated the discovery of subtype-specific molecular networks.

Conclusions:

• The developed statistical framework effectively detects network variations between biological subtypes.
• This approach facilitates the identification of novel molecular subgroups and their defining networks.
• The methods offer a powerful tool for network-based cancer subtyping and biomarker discovery.