disease genes [2] and drug discovery [3, 4]. Microarray is a form of
high-throughput genomic data providing relative measurements of
mRNA levels for thousands of genes in a biological sample. Besides
the gene expression research, there are gene-to-gene interaction
information inferred from microarray. And coexpression networks
are typically constructed from gene expression data using
correlation-based inference methods which have been commonly
used to reveal gene functions and investigate gene regulatory sys-
tems [5–7]. Based on the coexpression network, numerous meth-
ods emerged to identify the differential coexpression modules,
genes, or gene pairs which can further indicate the biological
mechanisms underlining the data [8–10]. In this chapter, we will
in detail introduce the coexpression network including the con-
struction of coexpression network and the differential coexpression
network analysis.2 From Gene Coexpression Network to Differential Coexpression Analysis
Microarray technology has provided a powerful approach for ana-
lyzing the genome-wide gene expression profiling [11]. In this
section, the construction of gene coexpression network will be
described. Then, the conception of gene differential coexpression
in coexpression network was defined and described in detail. Fig-
ure1 showed the workflow from gene coexpression network con-
struction to differential network analysis for expression data.2.1 Gene
Coexpression Network
Analysis
A gene coexpression network is an undirected graph, where the
graph nodes correspond to genes and edges between genes repre-
sent significant coexpression relationships. The network is usually
constructed by measuring the gene expression similarity, which
represents the coexpression relationships between genes. Gene
coexpression takes into account the gene-to-gene interactions and
makes it possible to investigate the whole-genome architecture
under a certain condition. And the Pearson correlation coefficient
is the most popular method to construct the gene coexpression
network. When construct the coexpression network, the pairwise
correlation should first be calculated. And then a correlation cutoff
should be given to filter the low-correlation pairs. There are two
thresholding strategies: the hard-thresholding method and the
soft-thresholding method. The first method includes the
correlation-based method [12], theq-value-based method [12],
the percent-based method [12], the rank-based method [13], and
other systematic threshold-finding methods [14]. The second
method needs a power value (β) to scale the correlation coefficient
and widen the difference between the low and high correlation
values [9].156 Bao-Hong Liu