1.Hard-Thresholding-Based Coexpression Network.
Value basedThere are mainly three types of value-based meth-
ods to construct the gene coexpression network including
correlation value-based, q value-based, and percent value-
based methods. Pearson correlation coefficients (PCCs) are
calculated based on the gene expression profiles. For a given
gene expression matrix withngenes andmsamples, all the
n(n1)/2 gene correlation pair number is generated. The
gene pairs with the correlation coefficient higher than theRth
or with the adjustedp-value for the correlation lower than the
qth will be left to form the coexpression network. On the other
hand, the absolute values for alln(n1)/2 gene correlation
pairs are sorted in decreasing order. A fraction (percent) of
gene pairs with the absolute correlation values will be retained.
Rank basedFirst calculate the Pearson correlation coefficient
(or some other similarity measure) between every pair of genes.
For each genegi, we rank all other genes by their similarity to
gi. And then connect every gene to thedgenes that are most
similar to it.
2.Soft-Thresholding-Based Coexpression Network.
WGCNA (weighted gene coexpression network analysis)
adopted the soft-thresholding strategy:xijand yijrepresent
the expression profile of genexand geney. First, calculate the
coexpression similarity measure of two genes by Pearson corre-
lation coefficientsij¼cor(xij,yij). Then the coexpression simi-
larity is transformed into the adjacency by raising the
coexpression similarity to a powerβ: aij¼sijβ, withβ1.
Different from the hard-thresholding method, the network
by soft-thresholding method is weighted and allows the adja-
cency to take on continuous values between 0 and 1.2.2 Coexpression
Network Comparison
and Differential
Coexpression Network
Analysis for Gene
Expression Data
Differential expression analysis considers each gene individually,
while their potential interactions are ignored. However, genes or
their protein products do not act in isolation; instead, they are
interacted with each other and act in close coordination. So differ-
ential coexpression analysis emerged to address this problem which
is based on the gene coexpression network analysis [15]. There are
three aspects for differential coexpression network analysis includ-
ing the topological characteristic comparison, differential coexpres-
sion gene module identification, and differential coexpression
genes and gene pair identification.äFig. 1(continued) thresholding method. Then for two network comparison, the network topological char-
acteristics were calculated. Next, the differential coexpression modules, differential coexpression genes, and
gene pairs were identified, and their function will be enriched by GO and KEGG analysis
158 Bao-Hong Liu