Target Discovery and Validation Reviews and Protocols

34 Imoto et al.
(4,5), and Bayesian networks (6–10). Although the paradigm based on microarray
technology with the clustering technique has made tremendous impacts on biomedical
research and practice, the strategy enhanced with computational gene network
analysis has not yet been well examined for practical applications. In this
chapter, we focus on a real application of the gene network analysis aimed at
understanding responses of the cellular system against a drug (11–13).
For identifying and validating drug target genes, the following three elements
are essential: (1) drug-affected genes, (2) druggable genes, and (3) drugactive
pathways. The definitions are as follows:
• Drug-affected genes are the candidates directly affected by the drug.
• Druggable genes regulate the drug-affected genes most strongly from upstream of
the gene network.
• Drug-active pathways are regulatory pathways perturbed by the drug.
These three types of information play an essential role in pharmacogenomics
and are unattainable from clustering methods, which are the standard for gene
expression analysis. This information can only be reached by elucidating the
gene network.
For identifying the drug-affected genes and druggable genes, Fig. 1 shows
the conceptual view of our strategy. From Fig. 1, it can be seen that the information
of gene networks plays an essential role. To create this information, we
use two types of microarray gene expression data: data obtained by single gene
disruptions, and data obtained by biological experiments of several dose and
time responses to the drug. We use two computational methods, virtual gene
technique and Bayesian networks, for extracting network information from
these data. The details are described in Subheading 3. The computer software
we used for efficient exploration of the estimated gene networks is in Note 1.
Because the drug-affected genes are candidate genes that are directly
affected by the drug, and the druggable genes potentially regulate the drugaffected
genes, this information is very important for identifying and validating
drug target genes. However, the drug-active pathway is not taken into account
in Fig. 1. Because the drug-active pathway plays an important role in understanding
the drug responses of the cellular system affected by the drug and the
prediction of the side effects, it is also a very important pathway for identifying
and validating drug target genes. The conceptual view of the computational
strategy for identifying drug-active pathways from the same microarray data in
Fig. 1 is shown in Fig. 2.
Our approach for identifying drug-active pathways involves mainly two
methods: (1) estimation of the gene network by using a Bayesian network
model from microarray data obtained by single gene disruptions, and (2)
identification of the drug-active pathways from time-course drug response
microarray data by the probabilistic inference on the gene network estimated