Yeast Systems Biology: Towards a Systems Understanding of Regulation of Eukaryotic Networks in Complex Diseases and Biotechnology - Systems Biology Concepts and Insights - page 332

databases and tools to analyze and visualize them are
available (e.g., Saccharomyces Genome Database, SGD;
www.yeastgenome.org/ , and GO Term Finder ( http://www.
yeastgenome.org/cgi-bin/GO/goTermFinder.pl ) ; Cytoscape
(open source bioinformatics software platform for visual-
izing molecular interaction networks www.cytoscape.org/
and BiNGO [113] ). Main biological networks are avail-
able via general databases (e.g., STRING, http://string-db.
org/ ) or specialized databases for a specific type of
network (e.g., DryGIN; http://drygin.ccbr.utoronto.ca/ ) .
In summary, annotated biological networks provide the
first background information in order to interpret the results
from high-throughput 'omics' studies using a combination
of tools (see above). These have been used in, for example,
basic functional annotation of unknown genes [114] ;to
investigate essential DNA damage repair mechanisms
[115] ; in systems biology studies on the control of cell
growth [53,108,116,117] and the cell cycle [118,119] ; and
in comparisons of different strains used in integrative
systems biology studies at the transcriptome, proteome and
metabolome levels [120] .

regulatory networks; [98] , protein interactome networks
[131] ; and of the whole yeast 'system' [93,132] ). At the
transcriptional level, clustering of genome-wide gene
expression profiles has been used to reveal the underlying
topology, inference of regulatory networks, and for the
functional annotation of unknown genes, based on the
'guilt-by-association' principle [133,134] . Dynamic tran-
script analysis has also been used to identify transcription
factors (TFs) and their target genes regulating transient
responses to stress [135] and specific nutritional perturba-
tions [93] .
Evidence of modularity of the yeast regulatory network,
identifying regulatory modules from gene expression data,
has also been demonstrated using Bayesian scoring and
decision trees [133] . Jothi and co-workers, with a graph
theoretical algorithm for hierarchical clustering of the yeast
regulatory network, and integrating heterogeneous data
sources with the network, revealed that TFs at the same
hierarchical level show similar dynamics [136] , and Youn
and co-workers developed a clustering and maximization
of likelihood-based learning algorithm from TF binding
and gene expression, which can identify condition-specific
regulation events [137] . As a final example, a study on
network topology in combination with studies of noise in
gene expression identified the tight transcriptional regula-
tion of genes with noisy expression [138] .
The main computational tools for inferring regulatory
networks are often developed in platforms such as
MATLAB (e.g., BayesNet; [139] or Bioinformatics
(Mathworks, Inc)), and R programming language
(e.g., LearnBayes; [140] and BoolNet [141] ). Standalone
tools are also available and, for example, yeast regulation
networks and high-throughput datasets have been used to
validate two of the publicly available computational tools
for integration of data with regulation networks (bio-
PIXIE, [142] ; ChIP-Array, [143] ; see below for a list of
tools developed for computation and visualization of
regulatory networks).
Inference and topological analysis of yeast regulation
networks is a growing area in systems biology. As more
experimental data from heterogeneous sources become
available, new refined global models of regulation networks
covering condition- and dynamics/time-dependence of the
networks will be more feasible. The following section,
focusing on physical and global genetic interaction
networks in yeast, provides a broader perspective to infer-
ence and topological analysis of regulatory networks.

Regulatory networks (e.g., signaling and regulatory
gene expression networks) (see also protein protein

e

and gene interaction networks)
Cellular systems regulate their activities via highly inter-
connected, dynamic, regulatory networks. Current knowl-
edge is based on information generated by biochemical
(bottom-up) approaches, although recent high-throughput
studies together with advanced computational methods
point to a larger number of new components (nodes) and
connections between them (edges) [13,100] .
The representation of a regulatory network is based on
few features, e.g., nodes' activity levels (discrete/contin-
uous); the relationships between nodes (directed/
non-directed/mathematical function/relation) and the
mathematical model implemented (stochastic/determin-
istic, or static/dynamic) [121] . Machine learning and
validation methods have been developed for network
inference in accordance with the above features [121

124] .
Integration of gene expression profiles with DNA-binding
profiles (e.g., ChIP-chip, [125] ; and ChIP-Seq, [126] )or
nucleosome positioning [127] and RNA-binding profiles
[128] are increasing the accuracy of network inference
methods considerably [129,130] .
The topology of yeast regulatory networks has been
investigated with different approaches that integrate high-
throughput data combined with a wide range of method-
ologies. Construction of gene expression correlation
networks as well as networks integrating gene expression,
protein interactions, growth phenotype data, and tran-
scription factor binding, revealed the modular organization
of the yeast regulatory networks (e.g., gene transcriptional

e

Protein protein and gene interaction networks

(PPI and GI networks)
Protein

e

protein interaction (PPI) networks describe phys-
ical interactions between proteins, taking place to mediate
the assembly of proteins into protein complexes, or e.g.,

e