Using Gene Expression Modeling to Determine Biological Relevance of Putative Regulatory Networks

Identifying gene regulatory networks from high-throughput gene expression data is one of the most important goals of bioinformatics, but it remains difficult to define what makes a ‘good’ network. Here we introduce Expression Modeling Networks (EMN), in which we propose that a ‘good’ regulatory network must be a functioning tool that predicts biological behavior. Interaction strengths between a regulator and target gene are calculated by fitting observed expression data to the EMN. ‘Better’ EMNs should have superior ability to model previously observed expression data. In this study, we generate regulatory networks by three methods using Bayesian network approach from an oxidative stress gene expression time course experiments. We show that better networks, identified by percentage of interactions between genes sharing at least one GO-Slim Biological Process terms, do indeed generate more predictive EMN’s.