Bayesian decomposition analysis of gene expression in yeast deletion mutants

Many methods have been proposed for the analysis of microarray data. In general, these methods are borrowed from statistics and data mining, and they ignore the underlying biology that gives rise to the data. Biological systems, such as cells, are complex, with constant activation and deactivation of multiple pathways in response to external and internal stimuli. Thus, of particular concern is the failure of many analysis methods to allow expression levels for a single gene to be explained as arising from multiple, different stimuli. Bayesian Decomposition, originally developed for spectral mixture analysis, overcomes this problem by permitting the discovered patterns within the expression data to overlap, allowing genes to belong to multiple groups. We present results of the application of Bayesian Decomposition to the deletion mutation data, demonstrating its ability to assign genes that are regulated by multiple pathways to multiple coexpression groups, allowing identification of changes to specific signalling pathways.