I will present a Bayesian hierarchical model that simultaneously performs clustering and feature selection. The model is combined with hidden Markov processes with three states for modeling functional dependence between features. The three states of the hidden Markov process allow us to obtain biologically meaningful clusters and to better discriminate them. Both simulation studies and gene expression analysis in a kidney cancer study illustrate the reliability and success of this method. We used Gene Ontology to define functional similarities between genes.