condition and uses an expectation-maximization algorithm to esti-
mate the posterior probability of each differential correlation
category.
Like DiffCorr, DGCA transforms correlation coefficients toz-
scores and uses differences inz-scores to calculate p-values of
differential correlation between genes. Like Discordant, DGCA
classifies differentially correlated gene pairs into the nine possible
categories. However, DGCA differs from the existing differential
correlation approaches. The simulation study indicated that the
DGCA performs better than the above DiffCorr, Discordant, and
EBcoexpress.3 Applications of Differential Coexpression Network Analysis in Cancer Research
Diseases caused by single gene’s variation can be detected by the
traditional differential expression analysis method and then the
molecular basis can be understood to discover disease biomarkers.
However, cancer is a complex disease caused by multiple genes’
aberration and can’t be caught by the above methods while differ-
ential coexpression network analysis takes full account of the inter-
actions of multiple genes and specific differential network of genes
and gene pairs which can be identified as dysfunctional in cancer by
comparing the difference of the coexpression networks. Deng et al.
constructed coexpression networks at the bladder cancer and nor-
mal state and found there were great differences between the two
networks in the network topological characteristics [30]. Since
cancer is caused by aberrations of multiple genes which possess
diverse functions and genes with similar functions are likely to be
coexpressed, Jia et al. identified lung cancer related modules in
coexpression networks using WGCNA and applied to facilitate
cancer research and clinical diagnosis [31]. Four modules of ovarian
cancer from a coexpression network were distinguished to be sig-
nificantly associated with biological processes such as cell cycle and
DNA replication [32]. Alexander et al. [33] explored gene net-
works in nine major human cancer types using a compendium of
publicly available data. The analysis resulted in a large collection of
high-resolution robust gene coexpression modules which offer
insight in cancer biology.
Besides the global and local changes for the coexpression net-
work under different cancer status, genes and gene pairs are also
differentially coexpressed [23, 33–37]. Li et al. identified 204 dif-
ferential coexpressed genes associated with cholangiocarcinoma
which provides a set of targets useful for future investigations into
molecular biomarker studies [37]. Fu et al. identified 37,094 dif-
ferentially coexpressed links (DCLs) and 251 DCGs and then con-
structed the regulatory network which enhanced theDifferential Coexpression Network Analysis for Gene Expression Data 163