Inference of Genetic Networks from Expression Profile by Graphical Gaussian Modeling

Abstract

Recent advance of DNA microarray technologies has made it possible to measure the expression levels of thousands of genes simultaneously, under different conditions. Elucidating patterns from the expression profile would provide us great insight into gene function and regulatory systems. For the purpose, several groups have developed the methods for clustering genes on the microarray. Here. clustering means partitioning the genes on a microarray into distinctive sets of genes that show similar expression patterns across the conditions. Hierarchical clustering [1], self-organizing mapping [2], and other clustering methods [3] have been applied to the expression profile data. Clustering genes with expression profiles can be utilized for the function prediction of gene products whose functions are unknown, and the identification of sets of genes whose expressions are regulated by the same mechanism. The important information underlying the expression profile data is the regulatory networks among genes. The expression level of a gene is regulated directly or indirectly by other genes. Here, we call the networks among genes the ‘genetic networks’. Inference of genetic networks from the expression profile would be valuable for the functional genomics. Modelings with the Boolean network [4], differential equations [5, 6], and the combination of the methods [7] have been investigated for the inference of the genetic networks. Graphical Gaussian modeling [8, 9] is a statistical method to infer and/or test a model for relationships among a plural of elements through representation of the model as a graph, when the elements are expressed as continuous variables. The central idea for the method is conditional independence, which tells us that partial correlation coefficient is appropriate to infer the relationships among elements rather than correlation coefficient. The method and the idea seemed to be applicable to the inference of the genetic networks from the expression profile data, because of the similarity in structure of inferred model. In this paper, we describe inference of the genetic networks from expression profiles by the graphical Gaussian modeling. This is the first report of the application of graphical modeling to the expression profile. We discuss how the method works to infer the genetic networks.