Gene module level analysis: identification to networks and dynamics.

Abstract

Nature exhibits modular design in biological systems. Gene module level analysis is based on this module concept, aiming to understand biological network design and systems behavior in disease and development by emphasizing on modules of genes rather than individual genes. Module level analysis has been extensively applied in genome wide level analysis, exploring the organization of biological systems from identifying modules to reconstructing module networks and analyzing module dynamics. Such module level perspective provides a high level representation of the regulatory scenario and design of biological systems, promising to revolutionize our view of systems biology, genetic engineering as well as disease mechanisms and molecular medicine.

Full text

Page 1

Available online at www.sciencedirect.com
Gene module level analysis: identification to networks and
dynamics
Xuewei Wang, Ertugrul Dalkic, Ming Wu and Christina Chan
Nature exhibits modular design in biological systems. Gene
module level analysis is based on this module concept, aiming
to understand biological network design and systems behavior
in disease and development by emphasizing on modules of
genes rather than individual genes. Module level analysis has
been extensively applied in genome wide level analysis,
exploring the organization of biological systems from
identifying modules to reconstructing module networks and
analyzing module dynamics. Such module level perspective
provides a high level representation of the regulatory scenario
anddesignofbiologicalsystems,promisingtorevolutionizeour
view of systemsbiology, genetic engineering aswell as disease
mechanisms and molecular medicine.
Address
Michigan State University, Chemical Engineering and Materials Science
Department, 2527 Engineering building, East Lansing, MI 48823, United
States
Corresponding author: Chan, Christina (krischan@egr.msu.edu)
Current Opinion in Biotechnology 2008, 19:482–491
This review comes from a themed issue on
Tissue, Cell and Pathyway Engineering
Edited by William Bentley and Michael Betenbaugh
Available online 3rd September 2008
0958-1669/$ – see front matter
# 2008 Elsevier Ltd. All rights reserved.
DOI 10.1016/j.copbio.2008.07.011
Introduction
Since the advent of molecular cell biology, researchers
have studied biological phenomenon mainly by analyzing
the function(s) of individual genes and proteins, and the
change(s) they exhibit in diseased states. This reduction-
ist approach helped discover many of the underlying
biological principles [1]. However, researchers sub-
sequently found that the relationship between genotype
and phenotype is too complicated to be ascribed to a
change in a single gene [2]. Traditional Mendel genetic
linkage test cannot explain complex diseases [3] and
crosstalk between the cell signaling cascades play a
significant role [4,5]. These observations, together with
the recent availability of‘omics’data, have revolutionized
the previous view of single gene–phenotype correlation
by demonstrating the importance of the inter-relation-
ships between genes. This has intensified the investi-
gation of protein function in the context of complex
biological systems and initiated a novel systematic
perspective of biological processes [6].
Modularity is now recognized as a design principle in
biological systems [7] and has been observed in protein–
protein interaction [8??], metabolic [9], and transcrip-
tional regulation networks [10]. Organized modularity
in the yeast protein–protein interaction network is evi-
dent in the different biological processes captured by
each functional subnetwork or module [8??]. Similarly,
metabolic and transcriptional networks organized into
topological units of different functions or regulation,
respectively, providing striking evidence of modulariza-
tion of biological systems [9]. The basic notion that
underlies the principle of such modularity lies in evol-
ution. Modularity may enhance the adaptability of bio-
logical systems to perform new functions [11], and
contribute to the systems’ robustness to perturbations
and their ability to maintain homeostasis [11,12]. For
example, network analysis has shown that the modular
structure of regulatory networks enables mutations to be
isolated to specific modules without affecting the overall
viability of the system [11–13]. Indeed, increasing evi-
dence suggests that a functional gene system is composed
of coordinated and interdependent modules [14?], and
such modularization is shaping current research in
biology.
With modularity being ubiquitous in biological systems,
applying module level analysis should aid the study of
biological systems at the different levels as well as the
systemdynamics.Genemodulelevelanalysisemphasizes
on groups or modules of genes rather than individual
genes [15–17]. Depending on their application, such
groups or modules could be cliques of physically inter-
acting proteins, clusters of gene expression profiles, path-
ways in a signaling, regulatory, or metabolic network, or
groups of manually pre-defined and functionally related
genes. Module level analysis has been extensively
applied in genome wide level analysis and systems
biology [18?,19]. For example, clustering has been used
frequently with microarray data to identify co-expressed
or co-regulated groups or modules of genes [18?,19].
Similarly, module-assisted methods have been used to
classify genes into coherent functional groups to predict
protein functions [18?,19]. Characteristics of a module
may include a higher number of interior connections
relative to interactions [20], similarity of function [20],
separability from its surrounding structure [21–23], or
Current Opinion in Biotechnology 2008, 19:482–491www.sciencedirect.com

Page 2

conservation of function [24]. With these characteristics,
modules qualify as building blocks of biological system;
thus, one can dissect the system’s structure and dynamics
by studying these modules in isolation, and then explore
the complex function of the whole system through the
interactions of the modules. To date, the focus of module
level analysis has been predominantly on identifying
individual modules and the qualitative descriptions of
the groupings of biological data. Lately, module level
analyses is shifting from a descriptive to a more quanti-
tative analysis of the inter-module relationships, that is,
studying their interplay through module network recon-
struction and dynamics, to understand the pathways,
mechanisms, and network regulations underlying human
diseases [25?]. This is driven by the need to mine and
integrate different large-scale datasets [26], requiring
quantitative measures to identify the modules and their
interactions or relationships [27??]. Gene module analysis
produces a high level representation of the regulatory
scenario and may help provide insight into biological
behaviors by analyzing the dynamic features of the gene
groups [23] or summarizing the network in the abstract
form of modules [27??].
Current gene module level analysis is focused predomi-
nantly on identifying the modules and several methods
have been applied, resulting in different approaches of
module analysis (Figure 1). In this review we explore the
methods and applications of gene module level analysis
from identifying modules, reconstructing module net-
works and analyzing module dynamics. Module identifi-
cation centers on the internal vertices and connections
within the modules, whereas module network and
dynamics focus on the external interactions between
the modules. Table 1 summarizes the strategies for
module level analysis discussed in this review.
Module identification
The first level of module analysis is to identify the gene
modules involved in specific biological processes or sys-
tems. Currently, the methods to identify gene modules
center around three major approaches: (1) network-based
Gene module level analysis: identification to networks and dynamics Wang et al.483
Figure 1
Schematic representation of gene module level analysis, from module identification to module network and dynamics. The first level of analysis is
module identification. The modules may be identified using different methods, including network-based, expression-based and pathway-based
approaches. On the basis of these approaches, different types of modules may be obtained (as indicated in the box with dash line), including
condense subnetworks, clusters of co-expressed genes and altered pathways. After identifying the modules, higher level of analysis including module
network reconstruction and module dynamic analysis may be performed on the modules. Module networks aim to represent the relationships between
modules in network form; such networks have been constructed on the basis of correlation between the expressions of the genes in the modules or
physical interactions between the genes in the modules. Module dynamics aim to analyze and predict the temporal behavior of modules; dynamic
models, such as state-space models, may be used for this purpose.
www.sciencedirect.com
Current Opinion in Biotechnology 2008, 19:482–491

Page 3

484
Tissue, Cell and Pathyway Engineering
Table 1
A summary of module level analysis strategies
Level of modular analysis Input dataMethodsOutputApplication examples
Module identification
Network-
based
approaches
Topological structure of
biological networks
Hierarchical clustering [7,28] Condense
subnetwork
Characterize modularity of biological
systems [31,32]
Predict protein functions [33,34]
Identify functional groups of
genes [35–38]
Graph clustering [29]
Clustering coefficient [30]
Additional information and
topological structures
Seed expansion [39]
Incorporating expression data [40]
Expression-
based
approaches
Static gene expressionTraditional clustering [41]Clusters of co-
expressed genes
Identify functional groups of genes
[41,43]
Identifying pathways activated under
specific conditions [44]
(hierarchical clustering, K-means)
Biclustering [42]
Model-based clustering [45,46]
Template-based matching [47,48]
Time-series expressions
Pathway-
based
approaches
Gene expressions in pathwaysEnrichment analysis [49–52,53?] Differentially
expressed pathways
Phenotype-relevant
pathways
Identify disease-related gene sets [54?];
and genetic alterations [55]
Discovering novel pathways [56,57]Supervised machine learning:
Nonparametric regression [58,59]
Discriminant analysis [60]
Partial least square regression [61]
Decision tree/random forests [62]
Module network construction
Physical interactions
(i.e. protein–protein)
Detecting the physical interactions
between modules [31]
Networks of modulesIdentifying pathway crosstalk [63]
Discovering coordinated
transcriptional modules [64]
Expressions of genes in modulesEigengene networks [65?]
Association graph [66??]
Module dynamics
Time-series expression dataDetecting dynamic changes of
modules and their connections
[54?,66??,67]
Difference between
static module
networks
Identifying targets for modulating
cell response [67]
Identifying pathways altered in
disease progression [54?,66??]
Describing and prediction module
dynamics [68?]
Control theory [68?]Dynamic models of
module behaviors
State-space model (SSM) [69?,70]
Current Opinion in Biotechnology 2008, 19:482–491
www.sciencedirect.com

Page 4

approaches identify highly connected subgraphs in the
biological networks as modules; (2) expression-based
approaches identify groups of co-expressed genes as
modules; and (3) prior pathways-based approaches identify
altered pathways as modules.
Network-based approaches
From a network perspective, modules can be defined as
subsets of vertices in the network with high intra-module
connectivity (Figure 2). Identification of modules from
biological networks focuses predominantly on protein
interaction networks.
Methods of finding modules from network topology use
mainly hierarchical and graph clustering [18?]. The hier-
archical clustering approach first classifies pairs of vertices
with a weight, and then clusters the vertices based on the
weights. Weights may be defined as the number of
vertex-independent paths, or the number of edge-inde-
pendent paths [7] (Figure 2). For example, a high number
of vertex-independent paths (i.e. paths between two
vertices involving different sets of intermediate vertices)
would correspond to a high weight, and have a higher
chance of being clustered together. Agglomerative algor-
ithms start with a complete set of vertices and add
individual edges based on the weights. Conversely, divi-
sive algorithms start with the whole network and proceed
by removing edges between vertices based on the
weights. After hierarchical transformation of the network
topology, qualitative or quantitative definitions of com-
munities are used to finally collect the modules. For
example, a typical quantitative definition requires that
each vertex in a community be more connected to ver-
tices within the community and less with those outside
[28] (Figure 2). In graph clustering, superparamagnetic
clustering and Monte Carlo optimization have been used
to identify highly connected clusters [29]. Alternatively,
Bader and Hogue used core clustering coefficients to
weight vertices and used these weights to find the dense
regions in a large protein interaction network [30].
Additional information, that is, gene expression, genetic
interactions, deletion phenotypes [18?], and metabolic
information [40] have been incorporated into the network
topology to identify modules. An example is to pre-define
groups of proteins previously known to interact as a
‘seed’. Members are added to the protein–protein inter-
action(PPI) network iftheysatisfyastatisticalconfidence
score. Statistical confidence is derived from distance
metrics obtained from multiple sources. Protein pairs
with short distances in networks derived from yeast
two-hybrid and mass spectrometry data are given a high
score, whereas the presence of a long distance lowers the
score [39]. Similarly, combining distance metrics from
gene expression data (based on high correlation coeffi-
cient) with distance metrics from metabolic interaction
networks (based on shortest path between two molecules
in the network) [40] through hierarchical clustering is
another method to identify modules (clusters).
Network based module identification is mainly applied to
large scale PPI data [31–35,71]. The identified modules
Gene module level analysis: identification to networks and dynamics Wang et al.485
Figure 2
A representative network with seven vertices (v1, v2, ... v7) and nine edges (e1, e2, ... e9). A path is a sequence of vertices and edges that start from
one vertex and end at another vertex and passes through each vertex only once. An example of a path between vertices v4 and v1 is a sequence of v4,
e3, e2, v1. However, this is not the only possible path between v4 and v1. Among the different possible paths between two vertices, vertex-
independent paths do not share any vertices except the first and the last vertices, whereas edge-independent paths do not share any edges. For
example, another path between v4 and v1 is a sequence of v4, e4, v1. Excluding the starting and ending vertices, this path does not share any edges
or vertices with the previous path (v4, e3, e2, v1); therefore these two paths are both vertex and edge-independent. Note all paths between v4 and v5
are both edge and vertex-dependent, because all paths have to pass through v1 and e1. Two different modules can be identified in this figure, with the
first module consisting of v1, v2, v3, v4 and the second module consisting of v5, v6, v7. The first module contains five edges and the second module
contains 3 edges, whereas there is only one edge between the modules. The figure was generated by using Cytoscape [100].
www.sciencedirect.com
Current Opinion in Biotechnology 2008, 19:482–491

Page 5

correlate well with protein complexes identified exper-
imentally and tend to contain proteins with similar func-
tions [32,35,71]. Thus, their success in identifying
modules encouraged their application in identifying nov-
el modules, pathways, and protein complexes [32,35,71].
For example, a protein of unknown function within the
same module as proteins of known function are deemed
functionally related. Likewise, these network-based
approaches have been applied to identify modules from
other biological networks, including metabolic [9], sig-
naling [72], and gene networks [37,38,73]. A disadvantage
of protein interaction network analysis is the possibly of
identifying modules that may not coexist in vivo [74].
Expression-based approaches
Network-based approaches identify modules from the
topological structure of the biological networks. By con-
trast, expression-based approaches use gene expression
data to infer modules of genes exhibiting similar expres-
sion patterns. Clustering methods are applied frequently
to identify modules of co-expressed genes. The motiv-
ation behind clustering is the assumption that co-
expressed genes are coordinately regulated [75], and
should reveal gene modules with similar functionality.
Traditional clustering methods, including agglomerative
and partitional techniques, are widely used to identify
modules of genes co-expressed across experimental con-
ditions [41,76]. Clustering of gene expressions is an active
area of research with several recent reviews [41,43].
Commonly used methods include hierarchical and K-
means clustering that are easy to implement and highly
efficient in discovering expression patterns across time or
conditions. Alternatively, biclustering method can cap-
ture such modules under a subset of experimental con-
ditions and time [42].
A limitation of clustering methods is their reliance on
functional interpretation of the identified modules
through a post hoc analysis, such as, enrichment analysis
ofthe gene categories. Togenerategene modulesthat are
more easily interpreted, different sources of information,
such as regulator–target gene relationships [77–79] and
pre-defined pathways [44], have been integrated with
expression data in the clustering procedure. Furthermore,
clustering methods do not take into account the temporal
dependencies of the expression data, limiting their ability
tocapturethedynamic behavior ofgenes[80].To address
thisissue,approachesbasedondynamicmodels[45,46]or
temporal trends of the gene expression data [47,48] have
been developed to identify modules from time-series
microarray data.
Prior pathways-based approaches
Pathway-based approaches aim to identify pathways,
namely, groups of genes, rather than individual genes
involved in biological processes. Single gene level
analyses typically use univariate statistical tests to detect
for genes with significant expression changes. However,
these analyses provide a long list of genes that oftentimes
are difficult to extract useful information and to put into
biologicalcontext.By
approaches detect subtle and coordinated changes in
expression of a group of genes in a pathway or with
related function [81]. Methods to identify the genes in
apathway includeenrichmentanalysis[49–52]andsuper-
vised machine learning approaches [58–62].
contrast,pathway-based
A number of enrichment analyses are used to identify
pathways that distinguish biological phenotypes based
upon function. Good reviews exist describing the prin-
ciples and issues [81,82] associated with the statistical
methodology. In particular, gene set enrichment analysis
(GSEA) [49] and similar methods [50–52] have been
applied to microarray data to unravel the mechanisms
of complex diseases [54?,55,56,83,84]. The gene expres-
sion profiles are viewed in terms of functional pathways,
and the pathways with a large number of differentially
expressed genes are identified as significant between two
sets of conditions. Recent enhancements allow handling
of more than two groups [53?,57]. Nevertheless, GSEA
evaluates one pathway at a time and does not capture the
pathway interdependences that contribute to changes in
the phenotypes. Furthermore, ontology mapping and
GSEA consider all the genes in a pathway equally import-
ant. But a recent study demonstrated that by including
information on the topology or position of the differen-
tially expressed genes in the pathway helped to identify
pathways that may be relevant to lung cancer but were
otherwise missed [85??]. Another concern is the quality of
thepathways,sincethepathwaysareusually curated from
the literature or other resources, non-relevant genes may
be included, or relevant genes may be excluded from the
pathways. Several attempts to minimize these misspeci-
fications include, defining signature genes to represent
pathway behavior [86], refining pathways to adapt to
specific conditions by removing unaltered genes from
the dataset [66??,87,88], and improving the functional
interpretation of gene groups by including additional
information (e.g. experimental conditions and biological
context) associated with the group [89].
Alternatively, supervised machine learning approaches
evaluate multiple pathways simultaneously, and thus
could account for pathway interdependence. Various
supervised machine learning methods, such as discrimi-
nant analysis [60], random forest [62], and partial least
square regression [61] have been applied to identify
relevant phenotype-specific pathways. The rationale is
if a pathway improves the prediction of the phenotype, its
relevance increases. These machine learning approaches
use pathways as input variables, and thus require a
quantitative representation of pathway activity. Certain
preprocessing methods have been applied to quantify
pathway activities by summarizing the gene expressions
486
Tissue, Cell and Pathyway Engineering
Current Opinion in Biotechnology 2008, 19:482–491www.sciencedirect.com

Page 6

in the pathways. For example, singular value decompo-
sition-based method [90] uses first principle component
to represent pathways, and shrunken centroid-based
methods [60,61] use centroids of the gene expressions
in the pathways, but information loss is a concern with
these preprocessing methods. To address the information
loss, an ensemble of classification models may be built in
lieu of predictive models based on single classifier. For
instance, a nonparametric pathway-based regression
(NPR) method [58,59] uses expression of the genes in
a pathway to characterize the activity of the pathway, and
the activity levels are regressed to the phenotypes to form
a predictive model. Similarly, Pang et al. [62] construct a
decision tree using the genes in a pathway and combine
the decision trees of all the pathways into a ‘forest’ to
predict the phenotype. These ensemble approaches pro-
vide a nice interpretation of the gene–pathway and path-
way–phenotype relationships.
reduced performance through the introduction of noise
ornon-relevantinformationwithinclusionofallthegenes
of the pathways [91]. To address the reduced perform-
ance, Edelman et al. formulated an enrichment score to
determine the variation in activity of only the differen-
tially expressed genes in a pathway among the individual
samples [53?].
Anotherconcern is
Module networks and dynamics
From a systems biology perspective, modules are build-
ing blocks that may be organized into a complex network
to achieve cellular regulation [92?]. Changes in the mod-
ular structure with respect to the inter-module connec-
tions are generally associated with alterations in the
properties and functions of the cell [22]. Investigation
of the inter-module relationships and dynamics enables
the prediction and manipulation and eventually our un-
derstanding of cellular behavior.
Module networks are constructed predominantly from
PPI and gene expression data. Early analyses of module
networks relied predominantly on PPI data [30,31]. After
identifying the modules in the network, the inter-module
interactions are generally analyzed through the role
played by the proteins that connect the different modules
[31]. For example, in studying the protein interaction
network of yeast filamentation, modules, that is, the
MAPK and HOG pathways were found to be connected
by the Ste11 gene. Indeed, support exists in the literature
to suggest Ste11 crosstalks between these two pathways
[93]. Thus far, these analyses are mostly limited to
identifying inter-module connections and lack a quanti-
tative description of module networks.
Using expression data to infer module networks is in its
early stages. One such attempt uses an eigengene net-
work approach, where hierarchical clustering of the
expression data is used in identifying the modules
[65?]. The eigengene, captured by the first principle
component of the gene expressions in a module, are used
in determining the associations between the modules.
The associations are characterized by the correlation
coefficients between the eiggenes, and modules with
high correlations are combined to construct the network
[65?]. Once the modules are represented by quantitative
measures, that is, the first principle component, available
approaches, such as information-theory approaches and
probabilistic models [94], can be used to infer the gene
module networks.
Attempts to overcome the limitations of single source
data have integrated expression with PPI data to con-
struct module networks [27??,95]. The incorporation of
physical interaction information, such as transcription
factor binding and PPI data, facilitate functional
interpretation of module associations. This approach
was able to infer co-regulation of functional modules in
Saccharomyces cerevisiae [64] and crosstalk between
modules in colorectal cancer metastasis [63]. Similarly,
incorporating pre-defined gene sets into the expression
analysis can take into account genes that are not differ-
entially expressed [54?,64,66??].
Dynamic analysis of module networks, thus far, focuses
on the changes within the modules and the connections
between the modules with time, as well as the dynamic
behaviors of the modules in the networks. To study the
temporal changes of themodule,a common approachisto
identify the modules, that is, the altered pathways, by
coupling time-series expression data with pathway infor-
mation, at each time point separately [54?,66??,67,96].
Some studies further analyze the connections between
the modules. By qualitatively analyzing the changes in
the module relationships across the different stages or
time points this approach has provided insight into tumor
progression [54?,66??] and response to acute heat stress
[96]. Dynamic models build upon time-series expression
data to analyze the dynamic behavior of the modules.
Quantitative measures adapted from dynamic system
theory, forexample,stability,
steady-state behavior, are applied to analyze the module
dynamics, by comparing the difference in these measures
between normal and abnormal cells [68?]. This approach
to exploring module dynamics, however, requires a cer-
tain degreeof mechanistic knowledgeof themodules(i.e.
equations that describe the dynamic regulations of the
genes), thus limiting their application to well-studied
pathways. In cases where such mechanistic knowledge
is not available, an alternative approach is to use state-
space model (SSM). SSM has been applied to infer
modules and dynamic regulation of yeast cell cycle and
cancer progression [69?,70]. In SSM, latent variables are
used to represent unobservable biological entities (i.e.
transcription factors or protein complexes), and the
temporal expression profiles are mapped to these latent
variables to model the dynamics and dependencies of
controllabilityand
Gene module level analysis: identification to networks and dynamics Wang et al.487
www.sciencedirect.com
Current Opinion in Biotechnology 2008, 19:482–491

Page 7

these latent variables. Genes that map to the same latent
variables are grouped together, and the dynamics and
dependencies of these latent variables are captured by
linear dynamic models used to describe the dynamic
behaviors of the modules and their interactions.
Conclusion
In the past three to four years, large biomolecular inter-
actionnetworks,constructedfromexperimentaldata,have
revealed many ‘emergent’ properties of network function-
ality [8??]. Concomitantly new problems have arisen,
namely how to model biological processes at different
organizational levels [97]. This has providedopportunities
as well as challenges in the development of network
biology approaches in uncovering the regulations and
mechanismsofcomplexdiseases[25?].Asdiscussedabove,
the property of modularity and module level analysis play
an important role in enabling the exploration of the regu-
lation and design of biological systems at a higher organ-
izational level. Therefore, recent progress in this field of
modulelevelanalysispromisestorevolutionizeourviewof
systems biology, genetic engineering as well as disease
mechanisms and molecular medicine. Nevertheless, chal-
lengesremain.Forthetimebeing,thequalityandcoverage
of the available data restrict our ability to perform module
level analyses that accurately recapitulate the biological
system.Toaddress thequalityofthedata suchaspathway
information, an edge ontology has been proposed to stan-
dardize pathway information [98]. An edge ontology will
help to distinguish different types of interactions by using
differentedgenotations.Sincethenumberofcomponents
may vary for the same pathway in different databases, a
well-defined edge ontology would provide less ambiguity
when comparing pathways across databases as well as
enable a more comprehensive representation of the
relationships within and among pathways. For coverage,
integrating available multi-source data can help improve
this situation [99]. Although, the amount and types of data
(i.e. continuous vs. discrete, static vs. dynamic) generated
are increasing, more computational methodologies are
needed to preprocess, analyze, integrate, and interpret
the various types of data in the pipeline. Finally, many
computational methods currently do not take full
advantageoftheavailabledataorapplystrongassumptions
that may not be realistic; for example, many dynamic
modelsassumetime-invariantstructuresmakingitdifficult
to capture the time-varying module relationships. As the
community strives to address these problems, we envision
that ultimately a comprehensive gene module level view
may accurately describe the biological system.
Acknowledgements
CC is supported partly by the National Institute of Health
(1R01GM079688, R21CA126136, R21RR024439), National Science
Foundation (BES0425821, DBI 0701709), MUCI Challenge Fund, and the
MSU Foundation on the Center for Systems Biology. ED is supported
partly by the Quantitative Biology and Modeling Initiative.
References and recommended reading
Papers of particular interest, published within the period of review,
have been highlighted as:
? of special interest
?? of outstanding interest
1.Nurse P: Reductionism and explanation in cell biology. Novartis
Found Symp 1998, 213:93-101 discussion 102–105.
2. Vladutiu GD: Heterozygosity: an expanding role in proteomics.
Mol Genet Metab 2001, 74:51-63.
3.Emahazion T, Feuk L, Jobs M, Sawyer SL, Fredman D, St Clair D,
Prince JA, Brookes AJ: SNP association studies in Alzheimer’s
disease highlight problems for complex disease analysis.
Trends Genet 2001, 17:407-413.
4. Pawson T, Saxton TM: Signaling networks—do all roads lead to
the same genes? Cell 1999, 97:675-678.
5. Dumont JE, Dremier S, Pirson I, Maenhaut C: Cross signaling,
cell specificity, and physiology. Am J Physiol Cell Physiol 2002,
283:C2-28.
6. Ge H, Walhout AJ, Vidal M: Integrating ‘omic’ information: a
bridge between genomics and systems biology. Trends Genet
2003, 19:551-560.
7. Girvan M, ME N: Community structure in social and biological
networks. Proc Natl Acad Sci U S A 2002, 99:7821-7826.
8.
??
A comprehensive review of network biology, from network properties to
biological significance and functional and mechanistic insights inferred
from analyzing the network. It introduces basic ideas, concepts, and
approaches in network biology and reviews the representative strategies
used to tackle the different biological problems from a top-down per-
spective. This paper is a good starting point in learning network biology.
HanJD:Understanding biologicalfunctionsthroughmolecular
networks. Cell Res 2008, 18:224-237.
9. Ravasz E, Somera AL, Mongru DA, Oltvai ZN, Barabasi AL:
Hierarchical organization of modularity in metabolic
networks. Science 2002, 297:1551-1555.
10. Ihmels J, Friedlander G, Bergmann S, Sarig O, Ziv Y, Barkai N:
Revealing modular organization in the yeast transcriptional
network. Nat Genet 2002, 31:370-377.
11. Yook SH, Oltvai ZN, Barabasi AL: Functional and topological
characterization of protein interaction networks. Proteomics
2004, 4:928-942.
12. Thieffry D, Romero D: The modularity of biological regulatory
networks. Biosystems 1999, 50:49-59.
13. Jeong H, Tombor B, Albert R, Oltvai ZN, Barabasi AL: The large-
scale organization of metabolic networks. Nature 2000,
407:651-654.
14.
?
Nature induced modularity into molecular networks to enhance their
adaptability to new functions. This review summarizes the evidence for
modularity, introduces recent models of modular evolution, and dis-
cusses the origin of modularity.
Wagner GP, Pavlicev M, Cheverud JM: The road to modularity.
Nat Rev Genet 2007, 8:921-931.
15. Chuang HY, Lee E, Liu YT, Lee D, Ideker T: Network-based
classification of breast cancer metastasis. Mol Syst Biol 2007,
3:140.
16. Von Dassow G, Odell GM: Design and constraints of the
Drosophilasegmentpolaritymodule:robustspatialpatterning
emerges from intertwined cell state switches. J Exp Zool 2002,
294:179-215.
17. von Dassow G, Meir E, Munro EM, Odell GM: The segment
polarity network is a robust developmental module. Nature
2000, 406:188-192.
18.
?
A review for the field of protein-function annotation. Besides providing an
overview of classical, network-based, and module-assisted methods, the
authors discuss recent developments in this area and compare the
Sharan R, Ulitsky IRS: Network-based prediction of protein
function. Mol Syst Biol 2007, 3:88.
488
Tissue, Cell and Pathyway Engineering
Current Opinion in Biotechnology 2008, 19:482–491www.sciencedirect.com

Page 8

performance of different methods in terms of accuracy and robustness.
The concept of the functional module is clearly defined and frequently
used in protein function annotation.
19. Wong DJ, Chang HY: Learning more from microarrays: insights
from modules and networks. J Invest Dermatol 2005,
125:175-182.
20. Alon U: Biological networks: the tinkerer as an engineer.
Science 2003, 301:1866-1867.
21. LipsonH,PollackJB,SuhNP:Ontheoriginofmodularvariation.
Evol Int J Org Evol 2002, 56:1549-1556.
22. Hartwell LH, Hopfield JJ, Leibler S, Murray AW: From molecular
to modular cell biology. Nature 1999, 402:C47-52.
23. Nordling TE, Hiroi N, Funahashi A, Kitano H: Deduction of
intracellular sub-systemsfromatopologicaldescription ofthe
network. Mol Biosyst 2007, 3:523-529.
24. Ettema T, van der Oost J, Huynen M: Modularity in the gain and
loss of genes: applications for function prediction. Trends
Genet 2001, 17:485-487.
25.
?
A comprehensive review of the applications of protein networks to
disease and medicine.
Ideker T, Sharan R: Protein networks in disease. Genome Res
2008, 18:644-652.
26. Curtis RK, Oresic M, Vidal-Puig A: Pathways to the analysis of
microarray data. Trends Biotechnol 2005, 23:429-435.
27.
??
Combines gene expression data with PPI networks to reveal a modular-
ized network view of the aging process, where the stochastic effects of
aging are linked to the dynamics of the modular network. To the best of
our knowledge, it is the first attempt at modeling a complex biological
process as the result of inter-module communications in a modular
network.Itmayalsobethefirstattemptatstudyingthemodulardynamics
on a large time-scale, within the time range of human life.
Xue H, Xian B, Dong D, Xia K, Zhu S, Zhang Z, Hou L, Zhang Q,
Zhang Y, Han JD: A modular network model of aging. Mol Syst
Biol 2007, 3:147.
28. Radicchi F, Castellano C, Cecconi F, Loreto VDP: Defining and
identifying communities in networks. Proc Natl Acad Sci U S A
2004, 101:2658-2663.
29. Spirin V, La M: Protein complexes and functional modules
in molecular networks. Proc Natl Acad Sci U S A 2003,
100:12123-12128.
30. Bader GD, Hogue CW: An automated method for finding
molecular complexes in large protein interaction networks.
BMC Bioinform 2003, 4:2.
31. Rives AW, Galitski T: Modular organization of cellular networks.
Proc Natl Acad Sci U S A 2003, 100:1128-1133.
32. Luo F, Yang Y, Chen C, Chang R, Zhou J, Scheuermann R:
Modular organization of protein interaction networks.
Bioinformatics 2007, 23:207-214.
33. Vazquez A, Flammini A, Maritan A, Vespignani A: Global protein
function prediction fromprotein–proteininteraction networks.
Nat Biotechnol 2003, 21:697-700.
34. Brun C, Herrmann C, Gue ´noche A: Clustering proteins from
interaction networks for the prediction of cellular functions.
BMC Bioinform 2004, 5:95.
35. Chen J, Yuan B: Detecting functional modules in the yeast
protein–protein interaction network. Bioinformatics 2006,
22:2283-2290.
36. Zhang S, Ning X, Zhang XS: Identification of functional modules
in a PPI network by clique percolation clustering. Comput Biol
Chem 2006, 30:445-451.
37. Aggarwal A, Guo DL, Hoshida Y, Yuen ST, Chu KM, So S,
Boussioutas A, Chen X, Bowtell D, Aburatani H et al.: Topological
and functional discovery in a gene coexpression meta-
network of gastric cancer. Cancer Res 2006, 66:232-241.
38. RuanXG,WangJL,LiJG:Anetworkpartitionalgorithmformining
gene functional modules of colon cancer from DNA microarray
data. Genomics Proteomics Bioinform 2006, 4:245-252.
39. Bader J: Greedily building protein networks with confidence.
Bioinformatics 2003, 19:1869-1874.
40. Hanisch D, Zien A, Zimmer R, Lengauer T: Co-clustering of
biological networks and gene expression data. Bioinformatics
2002, 18(Suppl. 1):S145-S154.
41. Kerr G, Ruskin HJ, Crane M, Doolan P: Techniques for clustering
gene expression data. Comput Biol Med 2008, 38:283-293.
42. Madeira SC, Oliveira AL: Biclustering algorithms for biological
data analysis: a survey. IEEE/ACM Trans Comput Biol Bioinform
2004, 1:24-45.
43. MillerDJ,Wang Y,KesidisG: Emergent unsupervised clustering
paradigms with potential application to bioinformatics. Front
Biosci 2008, 13:677-690.
44. Segal E, Friedman N, Koller D, Regev A: A module map showing
conditional activity of expression modules in cancer. Nat
Genet 2004, 36:1090-1098.
45. Ma P, Castillo-Davis CI, Zhong WX, Liu JS: A data-driven
clustering method for time course gene expression data.
Nucleic Acids Res 2006, 34:1261-1269.
46. Zeng YJ, Garcia-Frias J: A novel HMM-based clustering
algorithm for the analysis of gene expression time-course
data. Comput Stat Data Anal 2006, 50:2472-2494.
47. Ernst J, Nau GJ, Bar-Joseph Z: Clustering short time series
gene expression data. Bioinformatics 2005, 21:I159-I168.
48. Sacchi L, Bellazzi R, Larizza C, Magni P, Curk T, Petrovic U,
Zupan B: TA-clustering: cluster analysis of gene expression
profiles through temporal abstractions. Int J Med Inform 2005,
74:505-517.
49. Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL,
Gillette MA, Paulovich A, Pomeroy SL, Golub TR, Lander ES et al.:
Gene set enrichment analysis: a knowledge-based approach
for interpreting genome-wide expression profiles. Proc Natl
Acad Sci U S A 2005, 102:15545-15550.
50. Kim SY, Volsky DJ: PAGE: parametric analysis of gene set
enrichment. BMC Bioinform 2005, 6:144.
51. Klebanov L, Glazko G, Salzman P, Yakovlev A, Xiao Y: A
multivariate extension of the gene set enrichment analysis. J
Bioinform Comput Biol 2007, 5:1139-1153.
52. Jiang Z, Gentleman R: Extensions to gene set enrichment.
Bioinformatics 2007, 23:306-313.
53.
?
Edelman E, Porrello A, Guinney J, Balakumaran B, Bild A,
Febbo PG, Mukherjee S: Analysis of sample set enrichment
scores: assaying the enrichmentof sets ofgenes for individual
samples in genome-wide expression profiles. Bioinformatics
2006, 22:e108-e116.
Presents a method to quantify the activity of individual pathways in the
samples. This method may be very useful for inferring networks of
pathways. Once the pathways are quantified, computational methods
could be used to infer the network.
54.
?
Tomlins SA, Mehra R, Rhodes DR, Cao X, Wang L,
Dhanasekaran SM, Kalyana-Sundaram S, Wei JT, Rubin MA,
Pienta KJ et al.: Integrative molecular concept modeling of
prostate cancer progression. Nat Genet 2007, 39:41-51.
An impressive amount of gene sets (around 14 000) was collected, and
thus provides a useful resource for gene-set analysis.
55. Mullighan CG, Goorha S, Radtke I, Miller CB, Coustan-Smith E,
Dalton JD, Girtman K, Mathew S, Ma J, Pounds SB et al.:
Genome-wide analysis of genetic alterations in acute
lymphoblastic leukaemia. Nature 2007, 446:758-764.
56. Li Z, Srivastava S, Yang X, Mittal S, Norton P, Resau J, Haab B,
Chan C: A hierarchical approach employing metabolic and
gene expression profiles to identify the pathways that confer
cytotoxicity in HepG2 cells. BMC Syst Biol 2007, 1:21.
57. Tian L, Greenberg SA, Kong SW, Altschuler J, Kohane IS, Park PJ:
Discovering statistically significant pathways in expression
profilingstudies.ProcNatlAcadSciUSA2005,102:13544-13549.
58. Wei Z, Li H: Nonparametric pathway-based regression models
for analysis of genomic data. Biostatistics 2007, 8:265-284.
Gene module level analysis: identification to networks and dynamics Wang et al. 489
www.sciencedirect.com
Current Opinion in Biotechnology 2008, 19:482–491

Page 9

59. Luan Y, Li H: Group additive regression models for genomic
data analysis. Biostatistics 2008, 9:100-113.
60. Tai F, Pan W: Incorporating prior knowledge of gene functional
groups into regularized discriminant analysis of microarray
data. Bioinformatics 2007, 23:3170-3177.
61. Tai F, Pan W: Incorporating prior knowledge of predictors into
penalized classifiers with multiple penalty terms.
Bioinformatics 2007, 23:1775-1782.
62. Pang H, Lin A, Holford M, Enerson BE, Lu B, Lawton MP, Floyd E,
Zhao H: Pathway analysis using random forests classification
and regression. Bioinformatics 2006, 22:2028-2036.
63. Li Y, Agarwal P, Rajagopalan D: A global pathway crosstalk
network. Bioinformatics 2008, 24(12):1442-1447.
64. Petti AA, Church GM: A network of transcriptionally
coordinated functional modules in Saccharomyces cerevisiae.
Genome Res 2005, 15:1298-1306.
65.
?
Proposed a proof-of-concept approach to construct module network
from gene expression data, and a general framework to identify modules
and infer module networks.
Langfelder P, Horvath S: Eigengene networks for studying the
relationships between co-expression modules. BMC Syst Biol
2007, 1:54.
66.
??
Develops a data-driven pipeline to construct pathway networks from
condition-specificgeneexpressiondata.The pipelineincludes identifying
curated pathways relevant to the phenotypes, refining the curated path-
ways based on the expression data, and finally constructing networks of
the refined pathways.
EdelmanEJ,GuinneyJ,ChiJT,FebboPG,MukherjeeS:Modeling
cancer progression via pathway dependencies. PLoS Comput
Biol 2008, 4:e28.
67. Li Z, Srivastava S, Findlan R, Chan C: Using dynamic gene
module map analysis to identify targets that modulate
free fatty acid induced cytotoxicity. Biotechnol Prog 2008,
24:29-37.
68.
?
Characterizes the dynamic behavior of pathways by applying several
metricsfromcontrol theory.Pathway states areevaluatedinaholistic and
quantitative manner.
Xiong H, Choe Y: Dynamical pathway analysis. BMC Syst Biol
2008, 2:9.
69.
?
Hirose O, Yoshida R, Imoto S, Yamaguchi R, Higuchi T, Charnock-
Jones DS, Print C, Miyano S: Statistical inference of
transcriptional module-based gene networks from time
course gene expression profiles by using state space models.
Bioinformatics 2008, 24:932-942.
Provides a state-space model to infer module networks from short time-
series microarray data. Suggests a method to improve the parameter
estimation from limited time points.
70. Inoue LY, Neira M, Nelson C, Gleave M, Etzioni R: Cluster-based
network model for time-course gene expression data.
Biostatistics 2007, 8:507-525.
71. Zhang S, Ning X, Zhang X: Identification of functional modules
in a PPI network by clique percolation clustering. Comput Biol
Chem 2006, 30:445-451.
72. Baudot A, Angelelli J-B, Guenoche A, Jacq B, Brun C: Defining a
modular signalling network fromthefly interactome.BMCSyst
Biol 2008, 2:45.
73. Yip AM, Horvath S: Gene network interconnectedness and the
generalized topological overlap measure.BMC Bioinform 2007,
8:22.
74. Zhao SD, Li Y: Extracting functional modules from biological
pathways. Nat Preced 2007. hhttp://dx.doi.org/10.1038/
npre.2007.1457.1031i.
75. Boutros PC, Okey AB: Unsupervised pattern recognition: an
introduction to the whys and wherefores of clustering
microarray data. Brief Bioinform 2005, 6:331-343.
76. Zhao Y, Karypis G: Data clustering in life sciences. Mol
Biotechnol 2005, 31:55-80.
77. Segal E, Shapira M, Regev A, Pe’er D, Botstein D, Koller D,
Friedman N: Module networks: identifying regulatory modules
and their condition-specific regulators from gene expression
data. Nat Genet 2003, 34:166-176.
78. Ihmels J, Bergmann S, Barkai N: Defining transcription modules
using large-scale gene expression data. Bioinformatics 2004,
20:1993-2003.
79. Bar-Joseph Z, Gerber GK, Lee TI, Rinaldi NJ, Yoo JY, Robert F,
Gordon DB, Fraenkel E, Jaakkola TS, Young RA et al.:
Computational discovery of gene modules and regulatory
networks. Nat Biotechnol 2003, 21:1337-1342.
80. Bar-Joseph Z: Analyzing time series gene expression data.
Bioinformatics 2004, 20:2493-2503.
81. Nam D, Kim SY: Gene-set approach for expression pattern
analysis. Brief Bioinform 2008, 9(3):189-197.
82. Goeman JJ, Bu ¨hlmann P: Analyzing gene expression data in
terms of gene sets: methodological issues. Bioinformatics
2007, 23:980-987.
83. Baur JA, Pearson KJ, Price NL, Jamieson HA, Lerin C, Kalra A,
Prabhu VV, Allard JS, Lopez-Lluch G, Lewis K et al.: Resveratrol
improves health and survival of mice on a high-calorie diet.
Nature 2006, 444:337-342.
84. Lagouge M, Argmann C, Gerhart-Hines Z, Meziane H, Lerin C,
Daussin F, Messadeq N, Milne J, Lambert P, Elliott P et al.:
Resveratrol improves mitochondrial function and protects
against metabolic disease by activating SIRT1 and PGC-
1alpha. Cell 2006, 127:1109-1122.
85.
??
Draghici S, Khatri P, Tarca AL, Amin K, Done A, Voichita C,
Georgescu C, Romero R: A systems biology approach
for pathway level analysis. Genome Res 2007,
17:1537-1545.
Develops a novel method to identify altered pathways by integrating the
topological information of pathways, such as the positions and interac-
tions of the genes in the pathways. This method takes into account the
importance of the genes in the pathways based on the topological
information, unlike previous methods that consider the genes in a path-
way equally important.
86. Panteris E, Swift S, Payne A, Liu X: Mining pathway signatures
from microarray data and relevant biological knowledge. J
Biomed Inform 2007, 40:698-706.
87. Novak BA, Jain AN: Pathway recognition and augmentation by
computational analysis of microarray expression data.
Bioinformatics 2006, 22:233-241.
88. Gat-Viks I, Shamir R: Refinement and expansion of signaling
pathways: the osmotic response network in yeast. Genome
Res 2007, 17:358-367.
89. Hummel M, Meister R, Mansmann U: GlobalANCOVA:
exploration and assessment of gene group effects.
Bioinformatics 2008, 24:78-85.
90. Tomfohr J, Lu J, Kepler TB: Pathway level analysis of gene
expression using singular value decomposition. BMC
Bioinform 2005, 6:225.
91. Tsai CA, Chen CH, Lee TC, Ho IC, Yang UC, Chen JJ: Gene
selection for sample classifications in microarray
experiments. DNA Cell Biol 2004, 23:607-614.
92.
?
YanaiI,BaughLR,SmithJJ,RoehrigC,Shen-Orr SS,ClaggettJM,
Hill AA, Slonim DK, Hunter CP: Pairing of competitive and
topologically distinct regulatory modules enhances patterned
gene expression. Mol Syst Biol 2008, 4:163.
Provides experimental evidence that demonstrates molecular mechan-
isms attributed to coordination of the gene modules.
93. Cullen PJ, Schultz J, Horecka J, Stevenson BJ, Jigami Y, Sprague
GFJr:Defectsinproteinglycosylation cause SHO1-dependent
activation of a STE12 signaling pathway in yeast. Genetics
2000, 155:1005-1018.
94. Bansal M, Belcastro V, Ambesi-Impiombato A, di Bernardo D:
How to infer gene networks from expression profiles. Mol Syst
Biol 2007, 3:78.
95. Ulitsky I, Shamir R: Pathway redundancy and protein
essentiality revealed in the Saccharomyces cerevisiae
interaction networks. Mol Syst Biol 2007, 3:104.
490
Tissue, Cell and Pathyway Engineering
Current Opinion in Biotechnology 2008, 19:482–491www.sciencedirect.com

Page 10

96. Szustakowski JD, Kosinski PA, Marrese CA, Lee JH, Elliman SJ,
Nirmala N, Kemp DM: Dynamic resolution of functionally
related gene sets in response to acute heat stress. BMC Mol
Biol 2007, 8:46.
97. Schlitt T, Brazma A: Modelling gene networks at different
organisational levels. FEBS Lett 2005, 579:1859-1866.
98. Lu LJ, Sboner A, Huang YJ, Lu HX, Gianoulis TA, Yip KY,
Kim PM, Montelione GT, Gerstein MB: Comparing classical
pathways and modern networks: towards the development
of an edge ontology. Trends Biochem Sci 2007,
32:320-331.
99. Kutalik Z, Beckmann JS, Bergmann S: A modular approach
for integrative analysis of large-scale gene-expression
and drug-response data. Nat Biotechnol 2008, 26:531-539.
100. Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D,
Amin N, Schwikowski B, Ideker T: Cytoscape: a software
environmentforintegratedmodelsofbiomolecularinteraction
networks. Genome Res 2003, 13:2498-2504.
Gene module level analysis: identification to networks and dynamics Wang et al.491
www.sciencedirect.com
Current Opinion in Biotechnology 2008, 19:482–491