A Hierarchical Mixture Model for Gene Expression Data

We illustrate the use of a mixture of multivariate Normal distributions for clustering genes on the basis of Microarray data. We follow a hierarchical Bayesian approach and estimate the parameters of the mixture using Markov chain Monte Carlo (MCMC) techniques. The number of components (groups) is chosen on the basis of the Bayes factor, numerically evaluated using the Chib and Jelaizkov (2001) method. We also show how the proposed approach can be easily applied in recovering missing observations, which generally affect Microarray data sets. An application of the approach for clustering yeast genes according to their temporal profiles is illustrated.