Target Discovery and Validation Reviews and Protocols

Gene Networks 45
r is the dimension of , and ˆ is the mode of l λ ( | D). Based on the result of the
Laplace approximation, Imoto et al. (9) derived a criterion named Bayesian
network and nonparametric regression criterion (BNRC) for choosing the
optimal graph:
BNRC(G) =−2logπ prior (G)− r log(2π / n)
+ log | J λ ( ˆ
θ | D)|−2nl λ ( ˆ θ | D).
The optimal graph ˆG is chosen such that the criterion defined in Eq. 5 is
minimal. Imoto et al. (10) also extended the results of Imoto et al. (9) to handle
the nonparametric heteroscedastic regression.
Recently, research is underway to estimate gene networks from multisource
biological information (see Note 8). Also, the Bayesian networks force us to
consider gene networks as dags, whereas the real biological systems have cyclic
regulation. Some research has dealt with cyclic regulations based on the
dynamic models (see Note 9).
3.2.4. Application of Bayesian Networks for Identifying Druggable Genes
We selected 735 genes from the yeast genome for identifying drug targets
based on the 120 gene disruptant data and drug response time-course data.
These genes were selected based on the following strategy. We collected 314 genes
that are known as transcription factors. Ninety-eight of these 314 genes have
already been studied for their control mechanisms. The expression data for 735
genes chosen for our analysis includes the genes controlled by these 98 transcription
factors from 5871 genes in addition to nuclear receptor-like genes, which
have a pivotal role in gene expression regulation and are popular drug targets. We
have constructed the Bayesian network models described in Subheading 3.2.3.
of these 735 genes from 120 gene disruption conditions. The druggable genes
are the drug targets related to these drug-affected genes, which we want to identify
for the development of novel leads. We can explore the druggable genes
upstream of the drug-affected genes in the estimated gene network by the
Bayesian network method. Here, we focus on the nuclear receptor-like genes as
the druggable genes for two reasons. (1) In general, nuclear receptor proteins
are known to be useful drug targets, and together they represent more than 20%
of the targets for medications presently on the market. (2) Nuclear receptors are
involved in transcription regulatory affects that are directly measured in cDNA
microarray experiments.
Figure 7 shows a partial resulting network. The druggable genes (top)
that are the nuclear receptor-like genes in the circle connect directly to the
drug-affected genes, and other druggable genes have one intermediary gene per
(5)