Inference of Gene Regulatory Network by Bayesian Network Using Metropolis-Hastings Algorithm

Abstract

Bayesian networks are widely used to infer genes regulatory network from their transcriptional expression data. Bayesian network of the best score is usually chosen as genes regulatory model. However, without the hint from biological ground truth, and given a small number of transcriptional expression observations, the resulting Bayesian networks might not correspond to the real one. To deal with these two constrains, this paper proposes a stochastic approach to fit an existing hypothetical gene regulatory network, derived from biological evidence, with few available amount of transcriptional expression levels of the genes. The hypothetical gene regulatory network is set as an initial model of Bayesian network and fitted with transcriptional expression data by using Metropolis-Hastings algorithm. In this work, the transcriptional regulation of gene CYC1 by co-regulators HAP2 HAP3 HAP4 of yeast (Saccharomyces Cerevisiae) is considered as example. Due to the simulation results, ten probable gene regulatory networks which are similar to the given hypothetical model are obtained. This shows that Metropolis-Hastings algorithm can be used as a simulation model for gene regulatory network.