5.3 bnlearn Both the ARACNE and WGCNA software can only build up net-
works without directions. Such networks can be fast built and reveal
some global network features, such as scale-free or densely
connected modules, but they will not be able to tell which genes
are the upstream regulators and which genes are the downstream
targets. Therefore, after the genome-wide network is reconstructed
by ARACNE or WGCNA, important modules will be selected to
build refined regulatory networks with directions. Bayesian method
and software, such as bnlearn [67], are able to infer the small-scale
causal network in which the regulators and targets can be clearly
seen. But the computational complexity is too high; usually only a
dozen of genes can be handled. Therefore, the Bayesian method is
not suitable for genome-wide network reconstruction, despite its
network that is more informative. Usually, the genome-wide net-
work methods, such as ARACNE or WGCNA, are used first to get
the whole picture, and then the selected modules are further inves-
tigated using Bayesian method to get the refined regulatory
picture [68].
5.4 GeneReg GeneReg [69] is an R package for time delay gene regulatory
network construction from short-time course gene expression pro-
files. The basis is time delay linear regression model. Different from
ordinary linear regression model, this model has two parameters:
both time delay and the regulation coefficient. Time delay is the
time that the change of regulator’s gene expression is transmitted
and causes the change of target gene expression. It is difficult to
measure but extremely important for understanding basic
biological processes, such as cell cycle and signal cascade.
5.5 MCL MCL [33] is a cluster extraction software for graph analysis, which
can be utilized to identify modules of networks. As for the software,
it is an implementation of the Markov Cluster Algorithm, which is
based on simulation of stochastic flow in graphs. In practice, the
interaction matrix between genes of a network is used as input data
of the software, while modules of the network are produced as
output. In addition, some properties of a network, such as degree,
cluster coefficient, and betweenness of nodes, are presented also by
the software.
6 The Network Databases
The network databases store either experimentally validated regula-
tions/interactions or predicted regulations/interactions. They are
important for evaluating network reconstruction methods and
applying network analysis when there is not enough data to con-
struct the disease- or condition-specific network. Several widely
used network databases will be introduced.The Reconstruction and Analysis of Gene Regulatory Networks 149