A Bayesian approach for decision making on the identification of genes with different expression levels: an application to Escherichia coli bacterium data (Q764244)

Summary: A common interest in gene expression data analysis is to identify from a large pool of candidate genes the genes that present significant changes in expression levels between a treatment and a control biological condition. Usually, this is done using a statistical value and a cutoff value that are used to separate the genes differentially and non-differentially expressed. We propose a Bayesian approach to identify genes differentially expressed calculating sequentially credibility intervals from predictive densities which are constructed using the sampled mean treatment effects from all genes in the study excluding the treatment effects of genes previously identified with statistical evidence for difference. We compare our Bayesian approach with the standard ones based on the use of the t-test and modified t-tests via a simulation study, using small sample sizes which are common in gene expression data analysis. The results obtained report evidence that the proposed approach performs better than standard ones, especially for cases with mean differences and increases in treatment variance in relation to control variance. We also apply the methodologies to a well-known publicly available data set on Escherichia coli bacteria.