17.5 Gene Regulatory Networks:
The LegumeGRN Web Server
Gene expression datasets provide snapshots of
the transcriptomes of plant organs with/without
treatments and under various experimental con-
ditions. Genes and gene products interact with
each other in complex structured regulatory net-
works. Transcriptomes are valuable data to
uncover these complex regulatory interactions.
To predict gene interactions, several algorithms
have been developed using statistical and com-
putational tools. A Web-based computational
service was developed to build, test, analyze, and
visualize gene regulatory networks (GRNs)
(Wang et al. 2013 ). This Web server, called
LegumeGRN (http://legumegrn.noble.org), is
preloaded with Affymetrix GeneChip-based
transcriptomic data from Medicago, soybean, and
Lotus. The LegumeGRN Web server hosts the 83
Lotus transcriptomic experiments with the same
237 Affymetrix chips stored in the LjGEA and
described in21.4.
17.5.1 Lotus GRN Homepage
When users log in, they open the Lotus GRN
homepage by selecting Lotus in the “Submit
Gene Network prediction”tab. Then, users are
invited to (i) provide a list of probe sets/genes
that will be used to build the GRN, (ii) select
general options, (iii) select all or a subset of
preloaded transcriptomic data, and (iv) select one
or several GRN prediction algorithms
(Fig.17.4a).
The primary input of the LegumeGRN is a list
of Lotus probe sets/genes, which will serve to
build the gene regulatory network. This list of
genes has to be selected according to the purpose
of the study or according to the selected predic-
tive algorithm. As an example, according to the
purpose of the study, users may select a limited
number of genes such as organ-specific genes to
build a network related to a seed-specific mech-
anism by checking the specific experiments to
use for the GRN prediction. According to the
predictive algorithm, users may also select a
large set of genes when using“relevance net-
work” algorithm (i.e., co-expression network)
but are restricted to a limited number of genes
associated with a large number of transcriptomic
experiment when using graphical Gaussian
model for reasons of specificity and/or RAM
memory requirements specific to each algorithm.
Users will have access to six different robust
algorithms to predict GRNs (Marbach et al.
2012 ): relevance network based on Pearson’sor
Spearman correlation (Stuart et al. 2003 ),
graphical Guassian model (GGM, Schäfer and
Strimmer 2005 ), GENIE3 (Huynh-Thu et al.
2010 ), TIGRESS (Haury et al. 2012 ), CLR (Faith
et al. 2007 ) and parallel low-order PC algorithm
(Wang et al. 2010 ). Description of algorithms is
provided with their specificity by clicking on the
‘question mark’icons, and default settings are
proposed to users (Fig.17.4a). Finally, users may
change general options or keep default settings
concerning using transcription factors as main
connectors and the number of connections. By
default, immediate connections will be calculated
for transcription factors only (i.e., when“yes”is
checked for“using transcription factors”option)
and the number of edges is unlimited (i.e.,−1is
selected as a cutoff). The drop down menu
“submit gene network prediction”allows users to
enter their own in-house transcriptomic data that
are not contained in the Web server.17.5.2 Result PanelAfter calculation, GRN prediction results are
saved into LegumeGRN Web server users’
accounts, which allow users to store and keep
track of their analyses and results. Users have the
choice between downloading network results or
visualizing and analyzing them using an intuitive
Web-based GRN viewer (Fig.17.4b). The visu-
alization module consists of two parts: a graph-
ical output on the left panel and gene annotations
on the right panel (Fig.17.4b). The graphical17 A Tutorial onLotus japonicusTranscriptomic Tools 195