Bayesian Inference for Gene Expression and Proteomics: Computational Methods for Learning Bayesian Networks from High-Throughput Biological Data

Data from high-throughput technologies, such as gene expression microarrays, promise to yield insight into the nature of the cellular processes that have been disrupted by disease, thus improving our understanding of the disease and hastening the discovery of effective new treatments. Most of the analysis thus far has focused on identifying differential measurements, which form the basis of biomarker discovery. However, merely listing differentially expressed genes or gene products is not sufficient to explain the molecular basis of disease. Consequently, there is increasing interest in extracting more information from available data in the form of biologically meaningful relationships between the quantities being measured. The holy grail of such techniques is the robust identification of causal models of disease from data. The goal of this chapter is to survey computational learning methods that extract models of altered interactions that lead to and occur in the diseased state. Our focus is on methods that represent biological processes as Bayesian networks and that learn these networks from experimental measurements of cellular activity. Specifically, we will survey computational methods for learning Bayesian networks from highthroughput biological data.