Bayesian infinite mixture model based clustering of gene expression profiles.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Cluster analysis of gene expression data is crucial for understanding biological significance.
• Existing methods often require pre-specifying the number of clusters and may not fully account for data uncertainties.

Purpose of the Study:

• To develop a robust clustering procedure for gene expression profiles using Bayesian model-averaging.
• To address limitations of traditional clustering methods by enabling automatic determination of cluster numbers and incorporating experimental variability.

Main Methods:

• Developed a clustering procedure based on the Bayesian infinite mixture model.
• Utilized a Gibbs sampler to estimate the posterior distribution of clusterings.
• Summarized results using posterior pairwise probabilities of co-expression and complete linkage principle.

Main Results:

• Identified clusters of genes with similar expression patterns from the posterior distribution.
• The method automatically detects unclustered expression profiles and incorporates experimental replicates.
• Demonstrated the importance of including experimental variability in the clustering model.

Conclusions:

• The Bayesian model-averaging approach offers a significant advancement in model-based cluster analysis of gene expression data.
• This method provides a more accurate assessment of confidence in expression profile similarities by accounting for model selection uncertainties.