A general methodology for integration of microarray data

Abstract

We present a method for the integration of microarray datasets employing a fixed structure Bayesian network. Rather than learning all interactions simultaneously, we focus on undirected functional interactions between pairs of genes. Using Expectation Maximization, we learn one set of network parameters per functional category of interest. As we integrate further processing methods and refine the network structure, we hope both to improve performance and to increase the ability of the technique to expose specific biological properties of microarrays.