Deduction of a gene regulatory relationship framework from gene expression data by the application of graphical Gaussian modeling

Recently, we developed an automatic system for deducing a framework of regulatory relationships from gene expression data on a genomic scale. One of the merits of our system is that it simultaneously performs the gene classification and the relation inference from a large amount of gene expression data by combining the graphical Gaussian modeling with standard multivariate statistical techniques. In this paper, we describe our modifications of the previous system to estimate the cluster boundaries, by setting a user-defined threshold for measuring the linear relationship between the profiles of clusters. The modified system was applied to a whole set of yeast expression profiles of 6152 genes available at a web site, and the results obtained by the present analyses were statistically evaluated by a simulation that calculates the chance probability of the inferred framework of regulatory relationships, in addition to the statistical evaluation of the clusters by a previous method. In particular, the feasibility of the modified system is demonstrated by the comparison between two frameworks inferred from two sets of clusters that were estimated by distinctive thresholds. (C) 2002 Elsevier Science B.V. All rights reserved.