The full Bayesian significance test for mixture models: results in gene expression clustering

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Gene clustering groups genes with similar expression patterns, aiding in the discovery of biological relationships.
• Understanding gene expression is crucial for deciphering cellular processes and responses to various conditions.

Purpose of the Study:

• To develop and evaluate a new gene clustering method based on multivariate normal mixture models.
• To determine the optimal number of clusters using sequential hypothesis testing and a Bayesian approach.

Main Methods:

• Utilized multivariate normal mixture models for gene expression data clustering.
• Employed sequential hypothesis testing, starting with two components and increasing until acceptance, to predict the number of clusters.
• Applied the full Bayesian significance test, a Bayesian approach that avoids model complexity penalization.
• Validated the method on a cDNA microarray dataset of Saccharomyces cerevisiae gene expression across different strains.
• Assessed method sensitivity to data dimensionality using principal components analysis (PCA).

Main Results:

• The proposed Bayesian clustering method demonstrated consistent results in predicting the number of gene clusters.
• Comparison with Mclust (model-based clustering) indicated superior or comparable performance in terms of result consistency.
• The method effectively identified gene clusters from a dataset of 205 genes across 10 yeast strains.

Conclusions:

• The proposed Bayesian mixture model approach offers a robust and consistent method for gene clustering.
• This technique enhances the exploratory analysis of gene expression data, facilitating the identification of meaningful gene relationships.
• The sequential hypothesis testing framework provides an intuitive way to determine the appropriate number of clusters.