Constructing Gene Regulatory Network Based On Granger Causality

DNA microarray technology makes it possible to obtain a large amount of gene expression data quickly, which provides database for researchers. This technology promotes the study of relationships among genes greatly. Life activity is related to the expression of all genes in a cell. The expression of a gene is not isolated. The expression level of one gene may affected by one or more other genes and it has an effect on other genes at the same time. These regulatory relationships among genes constitute a gene regulatory network. Likewise, gene regulatory network manifests a complex regulation of many genes. The study of gene regulatory network helps us learn the relationship between one gene and other genes or its products. The analysis of disease gene regulatory networks instructs drug development, reduces abuse of drugs and provides valuable information for the diagnosis and treatment of complex diseases. Therefore, it is necessary to reconstruct and analyze gene regulatory networks. Gene regulatory network is renstructed by means of a variety of network models mostly, such as Boolean network model and Bayesian network models etc. However, the fact that gene data is time sequence is neglected, which is the essence of gene data.In order to improve deficiencies of traditional methods in constructing gene regulatory network, this paper puts forward a new associated study algorithm which is based on Granger causality. There are two presuppositions in this algorithm: firstly, gene expression series data as a kind of time series, analysis methods for time series is also applied to gene expression series data. Secondly, the relationship among genes is further understood by analyzing gene expression data. Granger causality is a useful tool for measuring variable causality, and we can use it to measure regulatory relationship among genes. This paper relates to many theories like autoregressive model and Granger causality. At first, dynamic characteristic hidden in gene expression series data is subtracted with autoregressive model. Then gene expression series data is projected into frequency domain to reduce the dimension of data and eliminate the influence of time on the causality. And then calculate the magnitude of Granger causality between genes. Finally, we establish a regulatory network contains all genes. To verify that autoregressive model is also applicable to gene expression series data, we use simulated yeast data and Escherichia coli data in theexperiment part. This algorithm performs better in precession rate and recall rate than the NIR algorithm in constructing gene network with real data of Saccharomyces cerevisiae.