A graphical chain model for inferring regulatory system networks from gene expression profiles

A procedure for graphical chain modeling has been designed for analyzing the expression profiles of genes that can be classified into several blocks in a natural order. Since the gene expression profiles often share similar patterns, the genes within a block are grouped into some clusters, as a prerequisite for the modeling. Then, the clusters in the naturally ordered blocks are regarded as variables. Finally, the associations of the variables within and between blocks are inferred by covariance selection in graphical Gaussian modeling. The newly designed procedure for graphical chain modeling was applied to 619 expression profiles of cell cycle related genes in yeast, which were selected from 792 genes experimentally identified as being transcribed in the order of four cell cycle phases, G1G1, SS, G2G2, and MM. By the application of the procedure, the 619 genes were classified into 50 clusters, and a chain graph was fitted for 50 clusters in the four phases. On focusing on the clusters, including transcription factors, characteristic relationships between the clusters emerged from the associations of the clusters within and between the four phases; one of the remarkable features is the distinctive relationships for the clusters between neighboring and non-neighboring phases. The merits and pitfalls of the graphical chain model are discussed in terms of its application to the field of molecular biology.