Modeling Large-Scale Gene Regulatory Networks using Gene Ontology-Based Clustering and Dynamic Bayesian Networks

For understanding the function of an organism, it is necessary to know "which", "how fast", and "when" the genes are expressed. A gene regulatory network represents how and when the genes interact with each other. Using genetic network modeling, it is possible to explain the cell functions at molecular level. DNA microarrays can measure the expression levels of thousands of genes simultaneously. Most of methodologies have proposed so far for modeling gene networks from microarray data take into account only a small number of genes. In this paper, a two steps method is proposed that can model large-scale Gene Regulatory Networks using time series microarray data. Firstly, genes are clustered based on existing biological knowledge (Gene Ontology annotations) and then a higher-order Markov dynamic Bayesian network is applied in order to model causal relationships between genes in each cluster. Finally the learned subnetworks are integrated to make a global network. This method is applied to reconstruct the regulatory network of 75 yeast genes from cell cycle gene expression dataset collected by Spellman et al. (1998). Comparing the results with the KEGG pathway map, indicates that this approach is capable of finding 31% of true relationships between genes (69% if directionality and time delay is not considered).