Goodness of fit: structural equation modeling methods to reconcile gene regulatory networks. Justin Fear^1,2, Daniel Campo³, Matthew Salomon³, Sergey Nuzhdin³, Lauren McIntyre². 1) Genetics & Genomics, University of Florida, Gainesville, FL; 2) Department of Molecular Genetics and Microbiology, University of Florida, Gainesville, FL; 3) Section of Molecular and Computational Biology, Department of Biological Sciences, University of Southern California, Los Angeles, CA.

   Gene regulatory networks (GRN) have moved to the forefront of methodologies providing insight into gene regulation and transcriptional response. By building GRNs from the bottom-up, decades molecular knowledge can be incorporated into network structure, while reducing the dimensionality of the problem. The majority of these molecular studies have used gene perturbation to painstakingly dissect regulatory networks, resulting in potential differences between hypothesized GRNs and those found in non-perturbed populations. How best to reconcile these differences and compare the topology of GRNs conducted under different experimental conditions is a complex problem, made more difficult by the number of possible comparisons among any pair of experiments. In order to address this problem a formal test for the goodness of fit is needed. Using structural equations modeling, we examine several goodness of fit statistics that have been previously developed in other fields, and test their application to GRNs. We demonstrate the utility of our approach using RNA-Seq data from a D. melanogaster heterozygous panel. Focusing on the InR/tor pathway, we compare the fit between virgin and mated female head tissue. We identified statistically significant differences in transcriptional regulation between these two experimental conditions. We also identified several gaps in the hypothesized network and searched for candidates to include in the GRN. This approach can be used to compare GRN topology between experimental conditions.