A semi-parametric Bayesian model for unsupervised differential co-expression analysis.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Differential co-expression analysis identifies dysregulated gene expression in disease by comparing gene co-expression across sample sets.
• Current methods focus on identifying genes co-expressed in one sample set but not another.

Purpose of the Study:

• To develop a novel probabilistic framework for jointly identifying sample groups (contexts) with specific co-expression patterns and gene groups with differing co-expression across these contexts.
• To enable unsupervised differential co-expression analysis for a broader range of applications.

Main Methods:

• A probabilistic framework was developed that uses gene co-clustering structure within samples as a similarity measure for grouping biological samples.
• This approach differs from traditional methods that rely on gene expression level similarities.
• The framework allows samples with discordant expression patterns to be grouped if their gene co-clustering structures are concordant.

Main Results:

• The method was applied to identify molecular subtypes of breast cancer, revealing reproducible differential co-expression patterns across independent datasets.
• Sample groupings derived from these patterns proved highly informative regarding disease outcome.
• Analysis of differentially co-expressed genes provided new insights into the estrogen receptor alpha (ERalpha) regulatory network.

Conclusions:

• The co-clustering structure serves as a valuable similarity measure for unsupervised analysis of gene expression profiles.
• This approach yields significant information about expression regulatory networks and their role in disease.