DOI: 10.1126/science.1069492
, 1662 (2002);295 Science
Hiroaki Kitano
Systems Biology: A Brief Overview
This copy is for your personal, non-commercial use only.
clicking here.colleagues, clients, or customers by
, you can order high-quality copies for yourIf you wish to distribute this article to others
here.following the guidelines
can be obtained byPermission to republish or repurpose articles or portions of articles
): January 27, 2011 www.sciencemag.org (this infomation is current as of
The following resources related to this article are available online at
http://www.sciencemag.org/content/295/5560/1662.full.html
version of this article at:
including high-resolution figures, can be found in the onlineUpdated information and services,
http://www.sciencemag.org/content/295/5560/1662.full.html#ref-list-1
, 11 of which can be accessed free:cites 18 articlesThis article
779 article(s) on the ISI Web of Sciencecited by This article has been
http://www.sciencemag.org/content/295/5560/1662.full.html#related-urls
100 articles hosted by HighWire Press; see:cited by This article has been
http://www.sciencemag.org/cgi/collection/cell_biol
Cell Biology
subject collections:This article appears in the following
registered trademark of AAAS.
is aScience2002 by the American Association for the Advancement of Science; all rights reserved. The title
CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005.
(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience
on January 27, 2011www.sciencemag.orgDownloaded from
Systems Biology: A Brief Overview
Hiroaki Kitano
To understand biology at the system level, we must examine the structure
and dynamics of cellular and organismal function, rather than the char-
acteristics of isolated parts of a cell or organism. Properties of systems,
such as robustness, emerge as central issues, and understanding these
properties may have an impact on the future of medicine. However, many
breakthroughs in experimental devices, advanced software, and analytical
methods are required before the achievements of systems biology can live
up to their much-touted potential.
Since the days of Norbert Weiner, system-level
understanding has been a recurrent theme in
biological science (1). The major reason it is
gaining renewed interest today is that progress
in molecular biology, particularly in genome
sequencing and high-throughput measure-
ments, enables us to collect comprehensive data
sets on system performance and gain informa-
tion on the underlying molecules. This was not
possible in the days of Weiner, when molecular
biology was still an emerging discipline. There
is now a golden opportunity for system-level
analysis to be grounded in molecular-level un-
derstanding, resulting in a continuous spectrum
of knowledge.
System-level understanding, the approach
advocated in systems biology (2), requires a
shift in our notion of “what to look for” in
biology. While an understanding of genes and
proteins continues to be important, the focus is
on understanding a system’s structure and dy-
namics. Because a system is not just an assem-
bly of genes and proteins, its properties cannot
be fully understood merely by drawing dia-
grams of their interconnections. Although such
a diagram represents an important first step, it is
analogous to a static roadmap, whereas what we
really seek to know are the traffic patterns, why
such traffic patterns emerge, and how we can
control them.
Identifying all the genes and proteins in an
organism is like listing all the parts of an
airplane. While such a list provides a catalog
of the individual components, by itself it is
not sufficient to understand the complexity
underlying the engineered object. We need to
know how these parts are assembled to form
the structure of the airplane. This is analo-
gous to drawing an exhaustive diagram of
gene-regulatory networks and their biochem-
ical interactions. Such diagrams provide lim-
ited knowledge of how changes to one part of
a system may affect other parts, but to under-
stand how a particular system functions, we
must first examine how the individual com-
ponents dynamically interact during opera-
tion. We must seek answers to questions such
as: What is the voltage on each signal line?
How are the signals encoded? How can we
stabilize the voltage against noise and exter-
nal fluctuations? And how do the circuits
react when a malfunction occurs in the sys-
tem? What are the design principles and pos-
sible circuit patterns, and how can we modify
them to improve system performance?
A system-level understanding of a biolog-
ical system can be derived from insight into
four key properties:
1) System structures. These include the net-
work of gene interactions and biochemical
pathways, as well as the mechanisms by which
such interactions modulate the physical proper-
ties of intracellular and multicellular structures.
2) System dynamics. How a system be-
haves over time under various conditions can
be understood through metabolic analysis,
sensitivity analysis, dynamic analysis meth-
ods such as phase portrait and bifurcation
analysis, and by identifying essential mecha-
nisms underlying specific behaviors. Bifurca-
tion analysis traces time-varying change(s) in
the state of the system in a multidimensional
space where each dimension represents a par-
ticular concentration of the biochemical fac-
tor involved.
3) The control method. Mechanisms that
systematically control the state of the cell can
be modulated to minimize malfunctions and
provide potential therapeutic targets for treat-
ment of disease.
4) The design method. Strategies to mod-
ify and construct biological systems having
desired properties can be devised based on
definite design principles and simulations,
instead of blind trial-and-error.
Progress in any of the above areas re-
quires breakthroughs in our understanding of
computational sciences, genomics, and mea-
surement technologies, and integration of
such discoveries with existing knowledge.
Identification of gene-regulatory logic (3)
and biochemical networks is a major challenge.
The conventional methods for creating a net-
work model include performing a series of ex-
periments to identify specific interactions and
conducting extensive literature surveys. Several
attempts are under way to create a large-scale,
comprehensive database on gene-regulatory
and biochemical networks (4). Although such
databases are useful sources of knowledge,
many network structures remain to be identi-
fied. Substantial research has been done on
expression profiling, in which clustering analy-
sis is used to identify genes that are coexpressed
with genes of known function (5, 6). Although
clustering analysis provides insight into the
“correlation” among genes and biological phe-
nomena, it does not reveal the “causality” of
regulatory relationships. Several methods have
been proposed to automatically discover regu-
latory relationships solely on the basis of mi-
croarray data (7–9). At present, such methods
use information derived from mRNA abun-
dance, so there is limited scope to infer causal-
ity based on transcriptional regulation. Posttran-
scriptional and posttranslational mechanisms of
regulation must be incorporated as large-scale
data become available, but many properties
have yet to be measured with sufficient accura-
cy or in high throughput. Although it is not
possible to incorporate all the desired data into
the automated discovery system, analysis of
transcriptional regulation may provide very
useful information because of the possible hy-
potheses it generates to allow us to infer the
network structure. In general, when multiple
hypotheses are generated by automated discov-
ery analysis, it reflects a lack of information.
This type of analysis can be combined with
entropy-based decision-making algorithms to
theoretically suggest an experiment that most
reduces the number of ambiguous network hy-
potheses. Although such algorithms have yet to
reach a level of practical application, they may
prove useful for determining the optimal order
of experiments needed to resolve ambiguous
hypotheses (10). Progress in this area would
lead to an increased emphasis on hypothesis-
driven research in biology (Fig. 1).
Once we have attained an understanding of
network structure, we will be able to investigate
network dynamics. In reality, analysis of dy-
namics and structure on the basis of network
dynamics are overlapping processes, because
dynamic analysis may yield useful predictions
of unknown interactions. For dynamic analysis
of a cellular system, we need to create a model.
But first it is important to carefully consider the
purpose of model building: Whether it is to
obtain an in-depth understanding of system be-
havior or to predict complex behaviors in re-
sponse to complex stimuli, we must first define
the scope and abstraction level of the model.
Sony Computer Science Laboratories, Inc., 3-14-13
Higashi-Gotanda, Shinagawa, Tokyo 141-0022, Japan,
and Kitano Symbiotic Systems Project, ERATO, JST,
and the Systems Biology Institute, Suite 6A, M31,
6-31-15 Jingumae, Shibuya, Tokyo 150-0001, Japan.
E-mail: kitano@csl.sony.co.jp
1 MARCH 2002 VOL 295 SCIENCE www.sciencemag.org1662
S YSTEMS B IOLOGY:THE G ENOME,LEGOME, AND B EYOND
REVIEW
on January 27, 2011www.sciencemag.orgDownloaded from
The choice of analytical method used de-
pends on the availability of biological knowl-
edge to incorporate into the model. A steady-
state analysis can be done using only the net-
work structure, without knowing the rate con-
stants for a particular reaction. For example,
flux balance analysis (FBA) was used to predict
switching of the metabolic pathway in Esche-
richia coli under different nutritional conditions
based on knowledge of only the metabolic net-
work structure; this was experimentally con-
firmed (11). With some knowledge of steady-
state rate constants, traditional stability analysis
and sensitivity analysis provide insights into
how systems behavior changes when stimuli
and rate constants are modified to reflect dy-
namic behavior. Bifurcation analysis, in which
a dynamic simulator is coupled with analysis
tools, can provide a detailed illustration of dy-
namic behavior (12, 13). This type of analysis
has become conventional in dynamic systems
and is already used in many studies on biolog-
ical simulation.
Once both the network structure and its
functional properties are understood for a large
number of regulatory circuits, studies on clas-
sifications and comparison of circuits will pro-
vide further insights into the richness of design
patterns used and how design patterns of regu-
latory circuits have been modified or conserved
through evolution. The hope is that intensive
investigation will reveal a possible evolutionary
family of circuits as well as a “periodic table”
for functional regulatory circuits.
Robustness is an essential property of bio-
logical systems (14). Understanding the mech-
anisms and principles underlying biological ro-
bustness is necessary for an in-depth under-
standing of biology at the system level. The
phenomenological properties exhibited by ro-
bust systems can be classified into three areas:
(i) adaptation, which denotes the ability to cope
with environmental changes; (ii) parameter in-
sensitivity, which indicates a system’s relative
insensitivity to specific kinetic parameters; and
(iii) graceful degradation, which reflects the
characteristic slow degradation of a system’s
functions after damage, rather than catastrophic
failure. In engineering systems, robustness is
attained by using (i) a form of system control
such as negative-feedback and feed-forward
control; (ii) redundancy, whereby multiple
components with equivalent functions are intro-
duced for backup; (iii) structural stability,
where intrinsic mechanisms are built to pro-
mote stability; and (iv) modularity, where sub-
systems are physically or functionally insulated
so that failure in one module does not spread to
other parts and lead to system-wide catastrophe.
Not surprisingly, these approaches used in en-
gineering systems are also found in biological
systems. Bacterial chemotaxis is an example of
negative feedback that attains all three aspects
of robustness (15–17). Redundancy is seen at
the gene level, where it functions in control of
the cell cycle and circadian rhythms, and at the
circuit level, where it operates in alternative
metabolic pathways in E. coli. Structural stabil-
ity provides insensitivity to parameter changes
in the network responsible for segment forma-
tion in Drosophila (18). And modularity is ex-
ploited at various scales, from the cell itself to
compartmentalized yet interacting signal-trans-
duction cascades (19).
To conduct a systems-level analysis, a com-
prehensive set of quantitative data is required.
Projects already under way, such as the Alli-
ance for Cellular Signaling (AfCS) (20), are
making large-scale measurements with the ul-
timate goal of creating an in-depth simulation
model of cells. Exploratory studies on modeling
should be done at the earliest stage of such a
project to identify where measurement bottle-
necks exist in building the final model and to
avoid acquiring data with little value for model
building, such as measurements of insufficient
coverage and accuracy.
Comprehensiveness in measurements re-
quires consideration of three aspects: (i) fac-
tor comprehensiveness, which reflects the
numbers of mRNA transcripts and proteins
that can be measured at once; (ii) time-line
comprehensiveness, which represents the
time frame within which measurements are
made; and (iii) item comprehensiveness,
which refers to the simultaneous measure-
ment of multiple items, such as mRNA and
protein concentrations, phosphorylation, lo-
calization, and so forth. Model-based exper-
iment planning dictates where accuracy is
critical and where it is not, so that resources
can be optimally allocated.
Complete system-level analysis of biolog-
ical regulation requires high throughput and
accurate measurements, goals that are per-
haps beyond the scope of current experimen-
tal practices. Technical innovations in exper-
imental devices, single-molecule measure-
ments, femto-lasers that permit visualization
of molecular interactions, and nano-technol-
ogies are critical aspects of systems biology
research. For example, microfluidic systems,
also known as micro-TAS (total analysis sys-
tem), enable minute quantities ( picoliters) of
samples to be measured more rapidly and
more precisely. Various prototypes for poly-
merase chain reaction and electrophoresis
have been developed (21–24). Such methods
not only speed up measurements, but also
encourage automation.
Software infrastructure is another critical
component of systems biology research. Al-
though attempts have been made to build sim-
ulation software and to make use of the many
analysis and computing packages originally de-
signed for general engineering purposes, there
is no common infrastructure or standard to en-
able integration of these resources. The Sys-
tems Biology Mark-up Language (SBML),
along with CellML, represent attempts to define
a standard for an XML-based computer-
Fig. 1. Hypothesis-driven
research in systems biol-
ogy. A cycle of research
begins with the selection
of contradictory issues of
biological significance
and the creation of a
model representing the
phenomenon. Models
can be created either au-
tomatically or manually.
The model represents a
computable set of as-
sumptions and hypothe-
ses that need to be test-
ed or supported experi-
mentally. Computational
“dry” experiments, such
as simulation, on models
reveal computational ad-
equacy of the assump-
tions and hypotheses
embedded in each mod-
el. Inadequate models
would expose inconsis-
tencies with established
experimental facts, and
thus need to be rejected
or modified. Models that
pass this test become subjects of a thorough system analysis where a number of predictions may be
made. A set of predictions that can distinguish a correct model among competing models is selected for
“wet” experiments. Successful experiments are those that eliminate inadequate models. Models that
survive this cycle are deemed to be consistent with existing experimental evidence. While this is an
idealized process of systems biology research, the hope is that advancement of research in computa-
tional science, analytical methods, technologies for measurements, and genomics will gradually trans-
form biological research to fit this cycle for a more systematic and hypothesis-driven science.
www.sciencemag.org SCIENCE VOL 295 1 MARCH 2002 1663
S YSTEMS B IOLOGY:THE G ENOME,LEGOME, AND B EYOND
on January 27, 2011www.sciencemag.orgDownloaded from
readable model definition that enables models
to be exchanged between software tools. Sys-
tems Biology Workbench (SBW) is built on
SBML and provides a framework of modular
open-source software for systems biology re-
search. Both SBML and SBW are collective
efforts of a number of research institutions shar-
ing the same vision (25).
How does the idea of systems biology im-
pact pharmaceutical industries and medical
practice? The most feasible application of sys-
tems biology research is to create a detailed
model of cell regulation, focused on particular
signal-transduction cascades and molecules to
provide system-level insights into mechanism-
based drug discovery (26–28). Such models
may help to identify feedback mechanisms that
offset the effects of drugs and predict systemic
side effects. It may even be possible to use a
multiple drug system to guide the state of mal-
functioning cells to the desired state with min-
imal side effects. Such a systemic response
cannot be rationally predicted without a model
of intracellular biochemical and genetic inter-
actions. It is not inconceivable that the U.S.
Food and Drug Administration may one day
mandate simulation-based screening of thera-
peutic agents, just as plans for all high-
rise building are required to undergo structural
dynamics analysis to confirm earthquake
resistance.
Although systems biology is in its infan-
cy, its potential benefits are enormous in both
scientific and practical terms. A transition is
occurring in biology from the molecular level
to the system level that promises to revolu-
tionize our understanding of complex biolog-
ical regulatory systems and to provide major
new opportunities for practical application of
such knowledge.
References and Notes
1. N. Weiner, Cybernetics or Control and Communica-
tion in the Animal and the Machine (MIT Press, Cam-
bridge, MA, 1948).
2. H. Kitano, Foundations of Systems Biology (MIT Press,
Cambridge, MA, 2001).
3. E. H. Davidson et al., Science 295, 1669 (2002).
4. Examples of such databases are: Signal Transduction
Knowledge Environment (STKE http://www.stke.org/);
KEGG (http://www.genome.ad.jp/); EcoCyc (http://
ecocyc.org/).
5. M. Eisen et al., Proc. Natl. Acad. Sci. U.S.A. 95, 14863
(1998).
6. S. Chu et al., Science 282, 699 (2000).
7. S. Onami et al.,inFoundations of Systems Biology,H.
Kitano, Ed. (MIT Press, Cambridge, MA, 2001), pp. 59 –75.
8. S. Imoto et al., Pacific Symposium on Biocomputing
2002 (World Scientific, Singapore, 2002), pp. 175–
186.
9. C. Yoo et al., Pacific Symposium on Biocomputing 2002
(World Scientific, Singapore, 2002), pp. 498 –509.
10. T. Ideker et al., Pacific Symposium on Biocomputing
(World Scientific, Singapore, 2000), pp. 302–313.
11. J. Edwards et al., Nature Biotechnol. 19, 125 (2001).
12. M. Borisuk et al., J. Theor. Biol. 195, 69 (1998).
13. K. Chen et al., Mol. Biol. Cell 11, 369 (2000).
14. M. E. Csete, J. C. Doyle, Science 295, 1664 (2002).
15. N. Barkai et al., Nature 387, 913 (1997).
16. U. Alon et al., Nature 397, 168 (1999).
17. T.-M. Yi et al., Proc. Natl. Acad. Sci. U.S.A. 97, 4649
(2000).
18. V. Dassaw et al., Nature 406, 188 (2000).
19. G. Weng et al., Science 284, 92 (1999).
20. Alliance for Cellular Signaling (http://www.cellular-
signaling.org/).
21. M. Burns et al., Science 282, 484 (1998).
22. R. Anderson et al., Nucleic Acids Res. 28, e60 (2000).
23. P. Simpson et al., Proc. Natl. Acad. Sci. U.S.A. 95,
2256 (1998).
24. P. Gilles et al., Nature Biotechnol. 17, 365 (1999).
25. Additional information can be obtained at http://
www.cds.caltech.edu/erato/ or http://www.sbml.org.
26. J. Gibbs, Science 287, 1969 (2000).
27. C. Sander, Science 287, 1977 (2000).
28. D. Noble, Science 295, 1678 (2002).
29. I thank J. Doyle, M. Simon, and members of ERATO
Kitano project for fruitful discussions. Supported by
the ERATO and BIRD program of the Japan Science
and Technology Corporation, and the Rice Genome
and Simulation Project of the Ministry of Agriculture,
Japan.
REVIEW
Reverse Engineering of Biological
Complexity
Marie E. Csete
1
and John C. Doyle
2
*
Advanced technologies and biology have extremely different physical
implementations, but they are far more alike in systems-level organization
than is widely appreciated. Convergent evolution in both domains pro-
duces modular architectures that are composed of elaborate hierarchies of
protocols and layers of feedback regulation, are driven by demand for
robustness to uncertain environments, and use often imprecise compo-
nents. This complexity may be largely hidden in idealized laboratory
settings and in normal operation, becoming conspicuous only when con-
tributing to rare cascading failures. These puzzling and paradoxical fea-
tures are neither accidental nor artificial, but derive from a deep and
necessary interplay between complexity and robustness, modularity, feed-
back, and fragility. This review describes insights from engineering theory
and practice that can shed some light on biological complexity.
The theory and practice of complex engineer-
ing systems have progressed so radically that
they often embody Arthur C. Clarke’s dictum,
“Any sufficiently advanced technology is in-
distinguishable from magic.” Systems-level
approaches in biology have a long history (1, 2)
but are just now receiving renewed mainstream
attention (3–13), whereas systems-level design
has consistently been at the core of modern
engineering, motivating its most sophisticat-
ed theories in controls, information, and com-
putation. The hidden nature of complexity
(“magic”) and discipline fragmentation with-
in engineering have been barriers to a dialog
with biology. A key starting point in devel-
oping a conceptual and theoretical bridge to
biology is robustness, the preservation of
particular characteristics despite uncertain-
ty in components or the environment (14).
Biologists and biophysicists new to study-
ing complex networks often express surprise at
a biological network’s apparent robustness
(15). They find that “perfect adaptation” and
homeostatic regulation are robust properties of
networks (16, 17), despite “exploratory mech-
anisms” that can seem gratuitously uncertain
(18–20). Some even conclude that these mech-
anisms and their resulting features seem absent
in engineering (20, 21). However, ironically, it
is in the nature of their robustness and complex-
ity that biology and advanced engineering are
most alike (22). Good design in both cases (e.g.,
cells and bodies, cars and airplanes) means that
users are largely unaware of hidden complexi-
ties, except through system failures. Further-
more, the robustness and fragility features of
complex systems are both shared and neces-
sary. Although the need for universal principles
of complexity and corresponding mathematical
tools is widely recognized (23), sharp differenc-
es arise as to what is fundamental about com-
plexity and what mathematics is needed (24).
This article sketches one possible view, using
experience and theoretical insights from engi-
neering complexity that are relevant to biology.
1
Departments of Anesthesiology and Cell and Devel-
opmental Biology, University of Michigan Medical
School, Ann Arbor, MI 48109, USA.
2
Control and
Dynamical Systems, Electrical Engineering, and Bio-
engineering, California Institute of Technology, Pasa-
dena, CA 91125, USA.
*To whom correspondence should be addressed. E-
mail: doyle@cds.caltech.edu
1 MARCH 2002 VOL 295 SCIENCE www.sciencemag.org1664
S YSTEMS B IOLOGY:THE G ENOME,LEGOME, AND B EYOND
on January 27, 2011www.sciencemag.orgDownloaded from