Structural Equation Modeling for Systems Biology

Issues, Controversies, Problems

Investigations of gene regulatory systems or complex functional networks among DNA, RNA, proteins and other cellular components in a living cell conventionally follow a standard protocol. After a DNA sequence is completed, the mRNA level is measured by a cDNA microarray, to reveal the gene expression profiles under various conditions. From this information, various algorithms, including Boolean and Bayesian networks, have been developed to infer complex gene networks (Akutsu et al., 2000; Friedman, et al., 2000). In our previous investigation, we developed an approach based on graphical Gaussian modeling (GGM) combined with hierarchical clustering (Aburatani et al., 2003). We can infer the huge network among all of the genes by the GGM approach, since this approach is suitable for massive amounts of gene expression data (Aburatani & Horimoto, 2005). However, GGM infers only the undirected graph, whereas the Boolean and Bayesian models infer the directed graph, which shows causality. Although all of these approaches are feasible for establishing relationships among genes, it is difficult to reveal the critical interactions between genes and the other cellular components, owing to the insufficient information about the other cellular components in the gene expression profiles. Since estimation of regulatory network between genes and the other cellular components is absolutely essential to uncover the mechanism of gene expression control, an alternative approach is needed.