Reverse Engineering of Gene Regulatory Networks Combining Dynamic Bayesian Networks and Prior Biological Knowledge

Abstract

An important problem in the Systems Biology field is the reverse engineering of gene regulatory networks from gene expression data. In this work, we addressed this problem using a probabilistic graphical model known as Dynamic Bayesian Network to model the regulatory relations among the genes. We also used a Boolean formalism, assuming that each gene can take on two possible values: 0 (not expressed) and 1 (expressed). To learn the Dynamic Bayesian Network from time-series gene expression data we search for the network structure that best matches the data using the Bayesian Information Criterion score and the BDe score and compared them. Besides that, we used a source of prior biological knowledge from a database named STRING, unlike most of the reverse engineering algorithms that does not take into account any source of additional information. The results show that this approach can improve the quality of the inferred networks, and we also showed that the Dynamic Bayesian Network performs better than its standard version, Bayesian Network.