Inferring Gene Regulatory Network From Dynamic Bayesian Network Based On Bootstrap Resampling

Dynamic Bayesian Network (DBN) model is one of the main models to construct a gene regulatory network. At present the data used to construct gene regulatory network is mainly gene expression data. Because of the flaws of the gene expression data and the lack of prior knowledge, it is difficult to achieve the desired quality standards. In order to better describe the regulatory relationships between genes and provide effective guidance and enough information for the researcher, it is very important to find an algorithm which can construct higher quality networks. The main purpose of this paper is to infer a highly reliable gene regulatory network based on the gene expression data. The research of this paper mainly concludes the resample of time series gene expression data and the construction of dynamic bayesian network based on the resample and network fusion technology.In this paper, the method of bootstrap resample about time series gene expression data was proposed. According to the characteristics of time series gene date, the traditional bootstrap resample algorithm is improved. With the local block bootstrap resample method, we can not only get the new data sets, but also hold the dependency of time points of the data. We construct the gene regulatory networks based on the bootstrap resample method. Use the new data set to establish dynamic bayesian network, and then fusion networks by a linear weighted fusion method based on a network score. This approach combines the advantages of all candidate networks and gives the idea of resample a better performance. At last we test the new algorithm on the yeast and Arabidopsis datasets. Sensitivity and specificity were used as an evaluation index. The results indicate that the new algorithm can find new regulation relationships which the original algorithm can not find. The new algorithm improves the reliability of the network.