Modeling multiple time units delayed gene regulatory network using dynamic Bayesian network

Time delay is an important biological feature of gene regulation, and it is widely observed by biological experiments. Most of the current applications which use dynamic Bayesian network to model gene regulatory network assume that the time delay between regulators and their targets is one time unit in a time series gene expression dataset. In fact, multiple time units delay is indicated to exist in the gene regulation process. In this thesis, a method of using higher-order Markov dynamic Bayesian network (HMDBN) to model multiple time units delayed gene regulatory network is proposed. A learning framework using mutual information and genetic algorithm is designed to learn the structure of a HMDBN from time series gene expression data. When applied to real-world yeast cell cycle gene expression datasets, the predicted gene regulatory networks are strongly supported by biological evidence and consistent with the yeast cell cycle phase information.