Systems Biology

springmin

Systems Biology: Toward System-level
Understanding of Biological Systems
Hiroaki Kitano
Systems biology is a new field in biology that aims at system-level understanding
of biological systems. While molecular biology has led to remarkable
progress in our understanding of biological systems, the current
focus is mainly on identification of genes and functions of their products
which are components of the system. The next major challenge is to
understand at the system level biological systems that are composed of
components revealed by molecular biology. This is not the first attempt at
system-level understanding, since it is a recurrent theme in the scientific
community. Nevertheless, it is the first time that we may be able to understand
biological systems grounded in the molecular level as a consistent
framework of knowledge. Now is a golden opportunity to uncover the essential
principles of biological systems and applications backed up by indepth
understanding of system behaviors. In order to grasp this opportunity,
it is essential to establish methodologies and techniques to enable us
to understand biological systems in their entirety by investigating: (1) the
structure of the systems, such as genes, metabolism, and signal transduction
networks and physical structures, (2) the dynamics of such systems,
(3) methods to control systems, and (4) methods to design and modify systems
for desired properties. This chapter gives an overview of the field of
systems biology that will provide a system-level understanding of life.
INTRODUCTION
The ultimate goal of biology is to understand every detail and principle of
biological systems. Almost fifty years ago,Watson and Crick identified the
structure of DNA (Watson and Crick, 1953), thus revolutionizing the way
biology is pursued. The beauty of their work was that they grounded biological
phenomena on a molecular basis. This made it possible to describe
every aspect of biology, such as heredity, development, disease, and evolution,
on a solid theoretical ground. Biology became part of a consistent
framework of knowledge based on fundamental laws of physics.
Since then, the field of molecular biology has emerged and enormous
progress has been made. Molecular biology enables us to understand biological
systems as molecular machines. Today, we have in-depth understanding
of elementary processes behind heredity, evolution, development,
and disease. Such mechanisms include replication, transcription,
translation, and so forth.
Large numbers of genes and the functions of their transcriptional
products have been identified, with the symbolic accomplishment of the
complete sequencing of DNA. DNA sequences have been fully identi-
fied for various organisms such as mycoplasma, Escherichia coli (E. coli),
Caenorhabditis elegans (C. elegans), Drosophila melanogaster, and Homo sapiens.
Methods to obtain extensive gene expression profiles are now available
that provide comprehensive measurement at the mRNA level. Measurement
of protein level and their interactions is also making progress
(Ito et al., 2000" Schwikowski et al., 2000). In parallel with such efforts,
various methods have been invented to disrupt the transcription of genes,
such as loss-of-function knockout of specific genes and RNA interference
(RNAi) that is particularly effective for C. elegans and is now being applied
for other species.
There is no doubt that our understanding of the molecular-level mechanisms
of biological systems will accelerate. Nevertheless, such knowledge
does not provide us with an understanding of biological systems as
systems. Genes and proteins are components of the system. While an understanding
of what constitutes the system is necessary for understanding
the system, it is not sufficient.
Systems biology is a new field of biology that aims to develop a
system-level understanding of biological systems (Kitano, 2000). Systemlevel
understanding requires a set of principles and methodologies that
links the behaviors of molecules to system characteristics and functions.
Ultimately, cells, organisms, and human beings will be described and
understood at the system level grounded on a consistent framework of
knowledge that is underpinned by the basic principles of physics.
It is not the first time that system-level understanding of biological
system has been pursued" it is a recurrent theme in the scientific community.
Norbert Wiener was one of the early proponents of system-level understanding
that led to the birth of cybernetics, or biological cybernetics
(Wiener, 1948). Ludwig von Bertalanffy proposed general system theory
(von Bertalanffy, 1968) in 1968 in an attempt to establish a general theory
of the system, but the theory was too abstract to be well grounded. A precursor
to such work can be found in the work of Cannon, who proposed
the concept of “homeostasis” (Cannon, 1933).With the limited availability
of knowledge from molecular biology, most such attempts have focused
on the description and analysis of biological systems at the physiological
level. The unique feature of systems biology that distinguishes it from
past attempts is that there are opportunities to ground system-level understanding
directly on the molecular level such as genes and proteins,
whereas past attempts have not been able to sufficiently connect systemlevel
description to molecular-level knowledge. Thus, although it is not
2 Hiroaki Kitano
the first time that system-level understanding has been pursued, it is the
first time to have an opportunity to understand biological systems within
the consistent framework of knowledge built up from the molecular level
to the system level.
The scope of systems biology is potentially very broad and different
sets of techniques may be deployed for each research target. It requires
collective efforts from multiple research areas, such as molecular biology,
high-precision measurement, computer science, control theory, and
other scientific and engineering fields. Research needs to be carried out
in four key areas: (1) genomics and other molecular biology research, (2)
computational studies, such as simulation, bioinformatics, and software
tools, (3) analysis of dynamics of the system, and (4) technologies for highprecision,
comprehensive measurements.
This constitutes a major multi-disciplinary research effort that will
enable us to understand biological systems as systems. But what does
this mean? “System” is an abstract concept in itself. It is basically an
assembly of components in a particular formation, yet it is more than a
mere collection of components. To understand the system, it is essential
that it can be not only to describe in detail, but also it to comprehend what
happens when certain stimuli or disruptions occur. Ultimately, we should
be able to design the system to meet specific functional properties. It takes
more than a simple in-depth description" it requires more active synthesis
to ensure that we have fully understood it.
To be more specific, in order to understand biological systems as systems,
we must accomplish the following.
System Structure Identification: First of all, the structures of the system
need to be identified, primarily such as regulatory relationships of
genes and interactions of proteins that provide signal transduction and
metabolism pathways, as well as the physical structures of organisms,
cells, organella, chromatin, and other components.
Both the topological relationship of the network of components as well
as parameters for each relation need to be identified. The use of highthroughput
DNA microarray, protein chips, RT-PCR, and other methods
to monitor biological processes in bulk is critical. Nevertheless, methods
to identify genes and metabolism networks from these data have yet to be
established.
Identification of gene regulatory networks1 for multicellular organisms is
even more complex as it involves extensive cell-cell communication and
physical configuration in three-dimensional space. Structure identification
for multicellular organisms inevitably involves not only identifying the
structure of gene regulatory networks and metabolism networks, but also
understanding the physical structures of whole animals precisely at the
1 In this article, the term “gene regulatory networks” is used to represent networks of gene
regulations, metabolic pathways, and signal transduction cascades.
3 Systems Biology: Toward System-level Understanding of Biological Systems
cellular level. Obviously, new instrumentation systems need to be developed
to collect necessary data.
System Behavior Analysis: Once a system structure is identified to a certain
degree, its behavior needs to be understood. Various analysis methods
can be used. For example, one may wish to know the sensitivity of
certain behaviors against external perturbations, and how quickly the system
returns to its normal state after the stimuli. Such an analysis not only
reveals system-level characteristics, but also provides important insights
for medical treatments by discovering cell response to certain chemicals
so that the effects can be maximized while lowering possible side effects.
System Control: In order to apply the insights obtained by system structure
and behavior understanding, research into establishing a method to
control the state of biological systems is needed. How can we transform
cells that are malfunctioning into healthy cells? How can we control cancer
cells to turn them into normal cells or cause apoptosis? Can we control
the differentiation status of a specific cell into a stem cell, and control it to
differentiate into the desired cell type? Technologies to accomplish such
control would enormously benefit human health.
System Design: Ultimately, we would like to establish technologies that
allow us to design biological systems with the aim of providing cures for
diseases. One futuristic example would be to actually design and grow
organs from the patient’s own tissue. Such an organ cloning technique
would be enormously useful for the treatment of diseases that require
organ transplants. There may be some engineering applications by using
biological materials for robotics or computation. By using materials that
have self-repair and self-sustaining capability, industrial systems will be
revolutionized.
This chapter discusses scientific and engineering issues to accomplish
in-depth understanding of the system.
MEASUREMENT TECHNOLOGIES AND EXPERIMENTAL METHODS
Toward Comprehensive Measurements
A comprehensive data set needs to be produced to grasp an entire picture
of the organism of interest. For example, the entire sequence has been deduced
for yeast, and a microarray that can measure the expression level
of all known genes is readily available. In addition, extensive studies of
protein-protein interactions using the two-hybrid method are being carried
out (Ito et al., 2000" Schwikowski et al., 2000). Efforts to obtain highresolution
spatiotemporal localization data for protein are underway.
C. elegans is an example of an intensively measured multi-cellular
organism. A complete cell lineage has already been identified (Sulston et
al., 1983" Sulston and Horvitz, 1977), the topology of the neural system
4 Hiroaki Kitano
has been fully described (White et al., 1986), the DNA sequence has been
fully identified (The C. elegans Sequencing Consortium, 1998), a project
for full description of gene expression patterns during development using
whole-mount in situ hybridization (Tabara et al., 1996) is underway, and
the construction of a systematic and exhaustive library of mutants has
begun. In addition, a series of new projects has started for measuring
neural activity in vivo, and for automatic construction of cell lineage in real
time using advanced image processing combined with special microscopy
(Yasuda et al., 1999" Onami et al., 2001a).
While yeast and C. elegans are examples of comprehensive and exhaustive
understanding of biological systems, similar efforts are now being
planned for a range of biological systems. Although these studies are currently
limited to understanding the components of the system and their
local relationship with other components, the combination of such exhaustive
experimental work and computational and theoretical research would
provide a viable foundation for systems biology.
Measurement for Systems Biology
Although efforts to systematically obtain comprehensive and accurate
data sets are underway, systems biology is much more demanding for
experimental biologists than the current practice of biology. It requires a
comprehensive body of data and control of the quality of data produced
so that it can be used as a reference point of simulation, modeling, and
system identification. Eventually, many of the current experimental procedures
must be automated to enable high-throughput experiments to be
carried out with precise control of quality. Needless to say, not all biological
experiments will be carried out in such an automated fashion, for
important contributions will be made by small-scale experiments. Nevertheless,
large-scale experiments will lay the foundation for system-level
understanding.
High-throughput, comprehensive, and accurate measurement is the
most essential part of biological science. While expectations are high for
a computational approach to overcome limitations in the traditional approach
in biology, it will never generate serious results without experimental
data upon which computational studies can be grounded. For the
computational and systems approach to be successful, measurement has
to be (1) comprehensive, (2) quantitatively accurate, and (3) systematic.
While the requirement for quantitative accuracy is obvious, the other
two criteria need further clarification. Comprehensiveness can be further
classified into three types:
Factor comprehensiveness: Comprehensiveness in terms of target factors
that are being measured, such as numbers of genes and proteins. It is important
that measurement is carried out intensively for the factors (genes
5 Systems Biology: Toward System-level Understanding of Biological Systems
and proteins) that are related to the central genes and proteins of interest.
Unless all genes and proteins are measured, how effectively measurement
covers the factors of interest is more important, rather than the sheer number
of factors measured.
Time-series comprehensiveness: In modeling and analysis of a dynamical
system, it is important to capture its behavior with fine-grain time series.
Traditional biological experiments tend to measure only the change
before and after a certain event. For computational analysis, data measured
at a constant time interval are essential in addition to traditional
sampling points.
Item comprehensiveness: There are cases where several features, such as
transcription level, protein interaction, phosphorylation, localization, and
other features, have to be measured intensively for the specific target.
“Systematic” means that measurement is performed in such a way that
obtained data can be consistently integrated. The ideal systematic measurement
is simultaneous measurement of multiple features for a single
sample. It is not sufficient to develop a sophisticated model and perform
analysis using only the mRNA or protein level. Multiple data need to be
integrated. Then, each data point has to be obtained using samples that are
consistent across various measurements. If samples are prepared in substantially
different ways, two data points cannot be integrated. Although
this requirement sounds obvious, very few data sets meet these criteria
today.
These criteria are elucidated in the scenario below with some examples
of requirements for experimental data.
For example, to infer genetic regulatory networks from an expression
profile, comprehensive measurement of the gene expression profile needs
to be carried out. Expression data in which only the wild-type is measured
is generally unusable for this purpose. The data should have a comprehensive
set of deletion mutant and overexpression of each gene. Desirable
data sets knock out all genes that are measured in the microarray. If only
a limited number of genes can be knocked out due to cost and time constraints,
it is critical that genes that are expected to be tightly coupled are
intensively knocked out rather than knocking out genes sparsely over the
whole possible regulatory network. This is due to computational characteristics
of the reverse engineering algorithm that constructs the gene
regulatory network from profile data. With such algorithms, sparse data
points leave almost unlimited ambiguities on possible network structures.
Even with the same number of data points, the algorithm produces much
more reliable network hypotheses if measured genes are closely related.
This is what is meant by factor comprehensiveness.
Time-series comprehensiveness is required for phenomena that are
time aligned. Time-series profile data need to be prepared with particular
caution in terms of time synchronization of samples to be measured.
6 Hiroaki Kitano
It is often the case in traditional experiments that only two measurement
points are set: one before the event and one after the event. For example,
many studies in cellular aging research measured the expression level of
aging-related genes for young cells, aged cells, and immortalized cells,
without measuring changes of expression level on fine-grain time series.
In some cases, time-series changes of expression level can be important
information to create candidate hypotheses or eliminate possible mechanisms.
In addition to measurements before and after a biologically interesting
event, measurement should be carried at a constant time interval.
Expression profile data that has reliable sample time synchrony and constant
time interval is most useful to enable the computational algorithm
to reliably fit models and parameters to experimental data.
Additional information from protein-protein interactions, such as
from yeast two-hybrid experiments, is very useful to infer protein-level
interactions that fill the gap between regulation of genes. Both protein interactions
and expression profiles should be measured on samples that are
prepared identically. This systematic measurement requirement is rather
hard to meet currently, because not many research groups are proficient
in multiple measurement techniques.
After obtaining gene regulatory networks, one needs to find out specific
parameters used in the network. To understand dynamics, it is essential
that each parameter regarding the network is obtained, so that various
numerical simulations and analyses can be performed. Such parameters
are binding constant, transcription rate, translation rate, chemical reaction
rate, degradation rate, diffusion rate, speed of active transport, etc. Except
for special cases, such as red blood cells, these constants are not readily
available. Measurement using extracts provides certain information, but
often these rate constants vary drastically in vivo. Ideally, comprehensive
measurement of major parameters would be performed in vivo, but any
measurement that gives reasonable estimates would be of great help. In
addition to parameter measurement, it is critically important to measure
the phosphorylation state at high resolution.
While accuracy is important, the level of accuracy required may vary
depending on which part of the system is to be measured. In some parts of
the network, the system behavior is sensitive to specific parameter values,
and thus has to be measured with high accuracy. In other parts of the
system, the system may be robust against fluctuations of large magnitude.
In such a case, it may often suffice to confirm that the parameter values fall
within the range of stability, instead of obtaining highly accurate figures.
The point is that not all parts of the system need to be tuned with the
same precision. For example, components for jet engines may have to
be produced with high precision, but seat belts do not have to achieve
the same precision as jet engine components. In future, the type and
accuracy requirements for experiments may be determined by theoretical
requirements.
7 Systems Biology: Toward System-level Understanding of Biological Systems
The examples given so far have focused on the process of identification
of network structure and parameters that enable simulation and analysis
of biochemical networks under the simplified assumption that all materials
are distributed homogeneously in the environment. Unfortunately, this
is not the case in biological systems. There are subcellular structures and
localization of transcription products that cause major diversion from a
naive model. Multi-cellular systems require measurement of cell-cell contact,
diffusion, cell lineage, gene expression during development, etc. For
accurate simulation and analysis, these features have to be measured in
a comprehensive, accurate, and systematic manner. We have not developed
devices to obtain high-throughput measurements for any of these
features. This is a serious issue that has to be addressed.
Next-generation Experimental Systems
To cope with increasing demands for comprehensive and accurate measurement,
a set of new technologies and instruments needs to be developed
that offers a higher level of automation and high-precision measurement.
First, dramatic progress in the level of automation of experimental
procedures for routine experiments is required in order to keep up
with increasing demands for modeling and system-level analysis. Highthroughput
experiments may turn into a labor-intensive nightmare unless
the level of automation is drastically improved. Further automation
of experimental procedures would greatly benefit the reliability of experiments,
throughput, and total cost of the whole operation in the long run.
Second, cutting-edge technologies such as micro-fluid systems, nanotechnology
and femto-chemistry may need to be introduced to design and
build next-generation experimental devices. The use of such technologies
will enable us to measure and observe the activities of genes and proteins
in a way that is not possible today. It may also drastically improve the
speed and accuracy of measurement for existing devices.
In those fields where there are obvious needs, such as sequencing
and proteomics, the above goals are already pursued. Beyond the development
of high-throughput sequencers using high-density capillary array
electrophoresis, efforts are being made to develop integrated microfabricated
devices that enable PCR and capillary electrophoresis in a single
micro device (Lagally et al., 1999" Simpson et al., 1998). Such devices
not only enable miniaturization and precision measurements, but will also
significantly increase the level of automation.
In the developmental biology of C. elegans, identification of cell lineage
is one of the major issues that needs to be accomplished to assist
analysis of the gene regulatory network for differentiation. The first attempt
to identify cell lineage was carried out entirely manually (Sulston
et al., 1983" Sulston and Horvitz, 1977), and it took several years to iden-
8 Hiroaki Kitano
tify the lineage of the wild type. Four-dimensional microscopy allowed
us to collect multi-layer confocal images at a constant time interval, but
lineage identification is not automatic. With the availability of exhaustive
RNAi knockout for C. elegans, high-throughput cell lineage identifi-
cation is essential to explore the utility of the exhaustive RNAi. Efforts are
underway to fully automate cell lineage identification, as well as threedimensional
nuclei position data acquisition (Onami et al., 2001a), fully
utilizing advanced image processing algorithms and massively parallel
supercomputers. Such devices meet some of the criteria presented earlier,
and provide comprehensive measurement of cell positions with high
accuracy. With automation, high-throughput data acquisition can be expected.
If the project succeeds, it can be used to automatically identify the
cell lineage of all RNAi knockout for early embryogenesis. The technology
may be augmented, but with major efforts, to automatically detect cell-cell
contact, protein localization, etc.
Combined with whole mount in situ hybridization and possible future
single-cell expression profiling, complete identification of the gene regulatory
network for C. elegans may be possible in the near future.
SYSTEM STRUCTURE IDENTIFICATION
There are various system structures that need to be identified, such as
the structural relationship among cells in the developmental process, detailed
cell-cell contact configuration, membrane, intra-cellular structures,
and gene regulatory networks. While each of these has significance in corresponding
research in systems biology, this section focuses on how the
structure of gene regulatory networks can be identified, primarily because
it is a subject of growing interest due to the rapid uncovering of genomic
information, and it is the control center of various cellular phenomena.
In order to understand a biological system, we must first identify
the structure of the system. For example, to identify a gene regulatory
network, one must identify all components of the network, the function of
each component, interactions, and all associated parameters. All possible
experimental data must be used to accomplish this non-trivial task. At
the same time, inference results from existing experiments should enable
the prediction of unknown genes and interactions, which can then be
experimentally verified.
The difficulty is that such a network cannot be automatically inferred
from experimental data based on some principles or universal rules, because
biological systems evolve through stochastic processes and are not
necessarily optimal. Also, there are multiple networks and parameter values
that behave quite similar to the target network. One must identify the
true network out of multiple candidates.
This process can be divided into two major tasks: (1) network structure
identification, and (2) parameter identification.
9 Systems Biology: Toward System-level Understanding of Biological Systems
Network Structure Identification
Several attempts have already been made to identify gene regulatory
networks from experimental data. They can be classified into two approaches.
BOTTOM-UP APPROACH
The bottom-up approach tries to construct a gene regulatory network
based on the compilation of independent experimental data, mostly
through literature searches and some specific experiments to obtain data
of very specific aspects of the network of interest. Some of the early attempts
of this approach are seen in the lambda phage decision circuit
(McAdams and Shapiro, 1995), early embryogenesis of Drosophila (Reinitz
et al., 1995" Hamahashi and Kitano, 1998" Kitano et al., 1997), leg formation
(Kyoda and Kitano, 1999a), wing formation (Kyoda and Kitano, 1999b),
eye formation on ommatidia clusters and R-cell differentiation (Morohashi
and Kitano, 1998), and a reaction-diffusion based eye formation
model (Ueda and Kitano, 1998). This approach is suitable when most of
the genes and their regulatory relationship are relatively well understood.
This approach is particularly suitable for the end-game scenario where
most of the pieces are known and one is trying to find the last few pieces.
In some cases, biochemical constants can be measured so that very precise
simulation can be performed. When most parameters are available, the
main purpose of the research is to build a precise simulation model which
can be used to analyze the dynamic properties of the system by changing
the parameters that cannot be done in the actual system, and to confirm
that available knowledge generates simulation results that are consistent
with available experimental data.
There are efforts to create databases that describe gene and metabolic
pathways from the literature. KEGG (Kanehisa and Goto, 2000) and Eco-
Cyc (Karp et al., 1999) are typical examples. Such databases are enormously
useful for modeling and simulation, but they must be accurate
and represented in such a way that simulation and analysis can be done
smoothly.
There have been some preliminary attempts to predict unknown genes
and their interactions (Morohashi and Kitano, 1998" Kyoda and Kitano,
1999a,b). These attempts manually searched possible unknown interactions
to obtain simulation results consistent with experimental data, and
did not perform exhaustive searches of all possible spaces of network
structures.
TOP-DOWN APPROACH
The top-down approach tries to make use of high-throughput data using
DNA microarray and other new measurement technologies. Already,
10 Hiroaki Kitano
there have been some attempts to infer groups of genes that have a tight
relationship based on DNA microarray data using clustering techniques
for the yeast cell cycle (Brown and Botstein, 1999" DeRisi et al., 1997"
Spellman et al., 1998) and development of mouse central neural systems
(D’haeseleer et al., 1999). Clustering methods are suitable for handling
large-scale profile data, but do not directly deduce the network structures.
Such methods only provide clusters of genes that are co-expressed in similar
temporal patterns. Often, easy-to-understand visualization is required
(Michaels et al., 1998).
Some heuristics must be imposed if we are to infer networks from such
methods. Alternative methods are now being developed to directly infer
network structures from expression profiles (Morohashi and Kitano, 1999"
Liang et al., 1999) and extensive gene disruption data (Akutsu et al., 1999"
Ideker et al., 2000). Most of the methods developed in the past translate
expression data into binary values, so that the computing cost can be
reduced. However, such methods seriously suffer from information loss
in the binary translation process, and cannot obtain the accurate network
structure. A method that can directly handle continuous-value expression
data was proposed (Kyoda et al., 2000b" Onami et al., 2001b) and reported
accurate performance without a serious increase in computational costs.
An extension of this method seems to be very promising for any serious
research on inference of gene regulatory networks.
Genetic programming has been applied to automatically reconstruct
pathways and parameters that fit experimental data (Koza et al., 2001).
The approach requires extensive computing power, and an example of
such is the 1,000 CPU cluster Beowulf-class supercomputer, but the approach
has the potential to be practical given the expected speed up of
processor chips.
Such extensions include the development of a hybrid method that
combines the bottom-up and the top-down approach. It is unlikely that
no knowledge is available before applying any inference methods" in
practical cases, it can be assumed that various genes and their interactions
are partially understood, and that it is necessary to identify the rest of the
network. By using knowledge that is sufficiently accurate, the possible
space of network structures is significantly reduced.
One major problem is that such methods cannot directly infer possible
modifications and translational control. Future research needs to address
integration of the data of the expression profile, protein-protein interactions,
and other experimental data.
Parameter Identification
It is important to identify only the structure of the network, but a set of
parameters, because all computational results have to be matched and
tested against actual experimental results. In addition, the identified net-
11 Systems Biology: Toward System-level Understanding of Biological Systems
work will be used for simulating a quantitative analysis of the system’s
response and behavioral profile.
In most cases, the parameter set has to be estimated based on experimental
data. Various parameter optimization methods, such as genetic
algorithms and simulated annealing, are used to find a set of parameters
that can generate simulation results consistent with experimental data
(Hamahashi and Kitano, 1999). In finding a parameter set, it must be noted
that there may be multiple parameter sets which generate simulation results
equally fitted to experimental data. An important feature of parameter
optimization algorithms used for this purpose is the capability to find
as many local minima (including a global minima) as possible, rather than
finding single global minima. This needs to be combined with a method
to indicate specific experiments to identify which one of such parameter
sets is the correct parameter set.
There are several methods to find optimal parameter sets such as
brute force exhaustive search, genetic algorithms, simulated annealing,
etc. Most of them are computationally expensive, and have not been considered
viable options in the past. But the situation has changed, and it
will change in future, too.
Although it is important to accurately measure and estimate the genuine
parameter values, in some cases parameters are not that critical. For
example, it was shown through an extensive simulation that the segment
polarity network in Drosophila exhibits a high level of robustness against
parameter change (von Dassow et al., 2000). For certain networks that are
essential for survival the networks need to be built robust against various
changes in parameters to cope with genetic variations and external
disturbances. For this kind of network, the essence is embedded into the
structure of the network, rather than specific parameters of the network.
This is particularly the case when feedback control is used to obtain robustness
of the circuits, as seen in bacterial chemotaxis (Yi et al., 2000).
Thus, parameter estimation and measurement may need to be combined
with theoretical analysis on sensitivity of certain parameters to
maintain functionalities of the circuit.
SYSTEM BEHAVIOR ANALYSIS
Once we understand the structures of the system, research will focus on
dynamic behaviors of the system. How does it adapt to changes in the
environment, such as nutrition, and various stimuli? How does it maintain
robustness against various potential damage to the system, such as
DNA damage and mutation? How do specific circuits exhibit functions
observed? To attain system-level understanding, it is essential to understand
the mechanisms behind (1) the robustness and stability of the system,
and (2) functionalities of the circuits.
It is not a trivial task to understand the behaviors of complex biolog-
12 Hiroaki Kitano
ical networks. Computer simulation and a set of theoretical analyses are
essential to provide in-depth understanding on the mechanisms behind
the circuits.
Simulation
Simulation of the behavior of gene and metabolism networks plays an important
role in systems biology research, and there are several ongoing
efforts on simulator development (Mendes and Kell, 1998" Tomita et al.,
1999" Kyoda et al., 2000a" Nagasaki et al., 1999). Due to the complexity of
the network behavior and large number of components involved, it is almost
impossible to intuitively understand the behaviors of such networks.
In addition, accurate simulation models are prerequisite for analyzing the
dynamics of the system by changing the parameters and structure of the
gene and metabolism networks. Although such analysis is necessary for
understanding the dynamics, these operations are not possible with actual
biological systems. Simulation is an essential tool not only for understanding
the behavior, but also for the design process. In the design of complex
engineering systems, various forms of simulation are used. It is unthinkable
today that any serious engineering systems could be designed and
built without simulation. VLSI design requires major design simulation,
thus creating one of the major markets for supercomputers. Commercial
aviation is another example. The Boeing 777 was designed based almost
entirely on simulation and digital prefabrication. Once we enter that stage
of designing and actively controlling biological systems, simulation will
be the core of the design process.
For simulation to be a viable methodology for the study of biological
systems, highly functional, accurate, and user-friendly simulator systems
need to be developed. Simulators and associated software systems often
require extensive computing power such that the system must run on
highly parallel cluster machines, such as the Beowulf PC cluster (Okuno
et al., 1999). Although there are some simulators, there is no system that
sufficiently covers the needs of a broad range of biology research. Such
simulators must be able to simulate gene expression, metabolism, and
signal transduction for a single and multiple cells. It must be able to
simulate both high concentration of proteins that can be described by
differential equations, and low concentration of proteins that need to be
handled by stochastic process simulation. Some efforts on simulating a
stochastic process (McAdams and Arkin, 1998) and integrating it with
high concentration level simulation are underway.
In some cases, the model requires not only gene regulatory networks
and metabolic networks, but also high-level structures of chromosomes,
such as heterochromatin structures. In the model of aging, some attempts
are being made to model heterochromatin dynamics (Kitano and Imai,
1998" Imai and Kitano, 1998). Nevertheless, how to model such dynamics
13 Systems Biology: Toward System-level Understanding of Biological Systems
and how to estimate the structure from sparse data and the current level
of understanding are major challenges.
The simulator needs to be coupled with parameter optimization tools,
a hypothesis generator, and a group of analysis tools. Nevertheless, algorithms
behind these software systems need to be designed precisely for
biological research. One example that has already been mentioned is that
the parameter optimizer needs to find as many local minima (including
global minima) as possible, because there are multiple possible solutions
of which only one is actually used. The assumption that the most optimal
solution is used in an actual system does not hold true in biological
systems. Most parameter optimization methods are designed to find the
global optima for engineering design and problem solving. While existing
algorithms provide a solid starting point, they must be modified to suit
biological research. Similar arguments apply to other software tools, too.
A set of software systems needs to be developed and integrated to
assist systems biology research. Such software includes:
• a database for storing experimental data,
• a cell and tissue simulator,
• parameter optimization software,
• bifurcation and systems analysis software,
• hypotheses generator and experiment planning advisor software, and
• data visualization software.
How these modules are related and used in an actual work flow is
illustrated in Figure 1.1. While many independent efforts are being made
on some of this software, so far only limited efforts have been made to
create a common platform that integrates these modules. Recently, a group
of researchers initiated a study to define a software platform for systems
biology. Although various issues need to be addressed for such a software
platform, the rest of this section describes some illustrative issues.
Efforts are being made to provide a common and versatile software
platform for systems biology research. The Systems Biology Workbench
project aims to provide a common middleware so that plug-in modules
can be added to form a uniform software environment.
Beside the software module itself, the exchange of data and the interface
between software modules is a critical issue in data-driven research
tools. Systems Biology Mark-up Language (SBML) is a versatile and common
open standard that enables the exchange of data and modeling information
among a wide variety of software systems (Hucka et al., 2000,
2001). It is an extension of XML, and is expected to become the industrial
and academic standard of the data and model exchange format.
Ultimately, a group of software tools needs to be used for disease modeling
and simulation of organ growth and control" this requires a comprehensive
and highly integrated simulation and analysis environment.
14 Hiroaki Kitano
Experimental Data
Database
Experimental Data
Interface
Measurement
Systems
Genome/Proteome
Database
Simulator
System Analysis Module
(Bifurcation analysis,
Flux Balance Analysis, etc.)
System Profile
Database
Hypotheses Generation
Experiment Planning
Module
Visualization Module
Parameter Optimization
Module
System Structure
Database
(A) Relationship among Software Tools
Simulator
Expression profile data
Two-hybrid data,
RT-PCR data, etc.
Parameter
optimizer
Hypotheses
generator
Gene regulation network
Metabolic cascade network
Signal transduction network
A set of plausible hypotheses
Predictions of genes and interactions
Experiment design
assistance system
Experiment plans
Biological
experiments
Dynamic systems analysis
Robustness, stability,
bifurcation, etc
Design pattern analysis
Design pattern extraction
(B) Workflow and software tools
Figure 1.1 Software tools for systems biology and their workflow
Analysis Methods
There have been several attempts to understand the dynamic properties
of systems using bifurcation analysis, metabolic control analysis, and sensitivity
analysis. For example, bifurcation analysis has been used to understand
the Xenopus cell cycle (Borisuk and Tyson, 1998). The analysis
creates a phase portrait based on a set of equations describing the essential
process of the Xenopus cell cycle. A phase portrait illustrates in which
operation point the system is acting, and how it changes behavior if some
of the system parameters are varied. By looking at the landscape of the
15 Systems Biology: Toward System-level Understanding of Biological Systems
phase portrait, a crude analysis of the robustness of the system can be
made.
A group of analysis methods such as flux balance analysis (FBA)
(Varma and Palsson, 1994" Edward and Palsson, 1999) and metabolic control
analysis (MCA) (Kacser and Burns, 1973" Heinrich and Rapoport,
1974" Fell, 1996) provides a useful method to understand system-level behaviors
of metabolic circuits under various environments and internal disruptions.
It has been demonstrated that such an analysis method can provide
knowledge on the capabilities of metabolic pathways that are consistent
with experimental data (Edward et al., 2001). While such methods are
currently aiming at analysis of the steady-state behaviors with linear approximation,
extention to dynamic and nonlinear analysis would certainly
provide a powerful tool for system-level analysis of metabolic circuits.
Several other analysis methods have already been developed for complex
engineering systems, particularly in the area of control dynamic systems.
One of the major challenges is to describe biological systems in the
language of control theory, so that we can abstract essential parts of the
system within the common language of biology and engineering.
ROBUSTNESS OF BIOLOGICAL SYSTEMS
Robustness is one of the essential features of biological systems. Understanding
the mechanism behind robustness is particularly important because
it provides in-depth understanding on how the system maintains its
functional properties against various disturbances. Specifically,we should
be able to understand how organisms respond to (1) changes in environment
(deprived nutrition level, chemical attractant, exposure to various
chemical agents that bind to receptors, temperature) and (2) internal failures
(DNA damage, genetic malfunctions in metabolic pathways). Obviously,
it is critically important to understand the intrinsic functions of the
system, if we are eventually to find cures for diseases.
Lessons from Complex Engineering Systems
There are interesting analogies between biological systems and engineering
systems. Both systems are designed incrementally through some sort
of evolutionary processes, and are generally suboptimal for the given task.
They also exhibit increased complexity to attain a higher level of robustness
and stability.
Consider an airplane as an example. If the atmospheric air flow is
stable and the airplane does not need to change course, altitude, or weight
balance, and does not need to take off and land, the airplane can be built
using only a handful of components. The first airplane built by theWright
brothers consisted of only a hundred or so components. The modern jet,
such as the Boeing 747, consists of millions of components. One of the
16 Hiroaki Kitano
major reasons for the increased complexity is to improve stability and
robustness. Is this also the case in biological systems?
Mycoplasma is the smallest self-sustaining organism and has only
about 400 genes. It can only live under specific conditions, and is very
vulnerable to environmental fluctuations. E. coli, on the other hand, has
over 4,000 genes and can live under varying environments. As E. coli
evolved it acquired genetic and biochemical circuits for various stress
responses and basic behavioral strategies such as chemotaxis (Alon et
al., 1999" Barkai and Leibler, 1997). These response circuits form a class
of negative feedback loop. Similar mechanisms exist even in eukaryotic
cells2.
A crude speculation is that further increases in complexity in multicellular
systems toward homo sapiens may add functionalities that can cope
with various situations in their respective ecological niche.
In engineering systems, robustness and stability are achieved by the
use of (1) system control, (2) redundancy, (3) modular design, and (4)
structural stability. The hypothesis is that the use of such an approach
is an intrinsic feature of complex systems, be they artificial or natural.
System Control: Various control schemes used in complex engineering
systems are also found in various aspects of biological systems. Feedforward
control and feedback control are two major control schemes, both
of which are found almost ubiquitously in biological systems. Feedforward
control is an open-loop control in which a set of predefined reaction
sequences is triggered by a certain stimulus. Feedback is a sophisticated
control system that closes the loop of the signal circuits to attain the desired
control of the system. A negative feedback system detects the difference
between desired output and actual output and compensates for such
difference by modulating the input. While there are feedforward control
methods, feedback control is more sophisticated and ensures proper control
of the system and it can be used with feedforward control. It is one
of the most widely used methods in engineering systems to increase the
stability and robustness of the system.
Redundancy: Redundancy is a widely used method to improve the system’s
robustness against damage to its components by using multiple
pathways to accomplish the function. Duplicated genes and genes with
similar functions are basic examples of redundancy. There is also circuitlevel
redundancy, such as multiple pathways of signal transduction and
metabolic circuits that can be functionally complementary under different
conditions.
Modular Design: Modular design prevents damage from spreading limitlessly,
and also improves ease of evolutionary upgrading of some of the
2 Discussion of similarity between complexity of engineering and biological systems as
described in this section was first made, as far as the author is aware, by John Doyle at
Caltech.
17 Systems Biology: Toward System-level Understanding of Biological Systems
Feedforward control
Controller Effector input output
Controller Effector input output
Feedback control
-
Figure 1.2 Feedforward control and feedback control
components. At the same time, a multi-functional module can help overcome
system failure in a critical part by using modules in other less critical
parts. Cellular systems are typical examples of modular systems.
Structural Stability: Some gene regulatory circuits are built to be stable
for a broad range of parameter variations and genetic polymorphisms.
Such circuits often incorporate multiple attractors, each of which corresponds
to functional state of the circuit" thus its functions are maintained
against change in parameters and genetic polymorphisms.
It is not clear whether such engineering wisdom is also the case in biological
systems. However, the hypothesis is that such features are somewhat
universal in all complex systems. It is conceivable that there are certain
differences due to the nature of the system it is built upon, as well as
the difference between engineering systems that are designed to exhibit
certain functions and natural systems that have reproduction as a single
goal where all functions are only evaluated in an integrated effect. Nevertheless,
it is worth investigating the univerality of principles. And, if there
are differences, what are they?
The rest of the section focuses on how three principles of robustness
exist also in biological systems. Of course, not all biological systems are
robust, and it is important to know which parts of the systems are not
robust and why. However, for this particular chapter, we will focus on
robustness of biological systems, because it is one of the most interesting
issues that we wish to understand.
Control
The use of explicit control scheme is an effective approach to improving
robustness. Feedforward control and feedback control are two major
methods of system control (Figure 1.2).
Feedforward control is an open-loop control in which a sequence of
predefined actions is triggered by a certain stimulus. This control method
18 Hiroaki Kitano
is the simplest method that works when possible situations and countermeasures
are highly predictable.
Feedback control, such as negative feedback, is a sophisticated control
method widely used in engineering. It feeds back the sign-inverted error
between the desired value and the actual value to the input, then the input
signal is modulated proportional to the amount of error. In its basic form,
it acts to minimize the output error value.
Feedback plays a major role in various aspects of biological processes,
such as E. coli chemotaxis and heat shock response, circadian rhythms, cell
cycle, and various aspects of development.
The most typical example is the integral feedback circuits involved in
bacterial chemotaxis. Bacteria demonstrates robust adaptation to a broad
range of chemical attractant concentrations, and so can always sense
changes in chemical concentration to determine its behavior. This is accomplished
by a circuit that involves a closed-loop feedback circuit (Alon
et al., 1999" Barkai and Leibler, 1997). As shown in Figure 1.3, ligands that
are involved in chemotaxis bind to a specific receptor MCP that forms a
stable complex with CheA and CheW. CheA phosphorylates CheB and
CheY. Phosphorylated CheB demethylates the MCP complex, and phosphorylated
CheY triggers tumbling behavior. It was shown through experiments
and simulation studies that this forms a feedback circuit which
enables adaptation to changes in ligand concentration. Specifically, for
any sudden change in the ligand concentration, the average activity level
that is characterized by the tumbling frequency quickly converges to the
steady-state value. This means that the system only detects acute changes
of the ligand concentration that can be exploited to determine tumbling
frequency, but is insensitive to the absolute value of ligand concentration.
Therefore, the system can detect and control its behavior to move to a
high attractant concentration area in the field regardless of the absolute
concentration level without saturating its sensory system. Detailed analysis
revealed that this circuit functions as an integral feedback (Yi et al.,
2000) —the most typical automatic control strategy.
In bacteria, there are many examples of sophisticated control embedded
in the system. The circuit that copes with heat shock, for example, is
a beautiful example of the combined use of feedforward control and feedback
control (Figure 1.4). Upon heat shock, proteins in E. coli can no longer
maintain their normal folding structures. The goal of the control system is
to repair misfolding proteins by activating a heat shock protein (hsp), or
to dissociate misfolding proteins by protease. As soon as heat shock is imposed,
a quick translational modulation facilitates the production of σ32
factor by affecting the three-dimensional structure of rpoH mRNA that
encodes σ32. This leads to the formation of σ32-RNAP holo-enzyme that
activates hsp that repair misfolded proteins. This process is feedforward
control that pre-encodes the relationship between heat shock and proper
course of reactions. In this process, there is no detection of misfolded pro-
19 Systems Biology: Toward System-level Understanding of Biological Systems
MCP
CheW
CheA
MCP
CheW
CheA
CheR
m
CheB
CheY
CheZ
CheB P
CheY P
Figure 1.3 Bacterial chemotaxis related feedback loop
teins to adjust the translational activity of σ32. Independently, DnaK and
DnaJ detect misfolded proteins and release σ32 factor, that has been bound
with DnaK and DnaJ. Free σ32 activates transcription of hsp, so that misfolded
proteins are repaired. This process is negative feedback control,
because the level of misfolded proteins is monitored and it controls the
activity of σ32 factor.
Another example demonstrating the critical role of the feedback system
is seen in growth control of human cells. Growth control is one of the
most critical parts of cellular functions. The feedback circuit involved in
p53 presents a clear example of how feedback is used (Figure 1.5). When
DNA is damaged, DNA-dependent kinase DNA-PK is activated. Also,
ATM is phosphorylated, which makes ATM itself in an active state and
promotes phosphorylation of the specific locus of the p53 protein. When
this locus is phosphorylated, p53 no longer forms a complex with MDM2,
and escapes from dissociation. The phosphorylation locus depends on
what kind of stress is imposed on DNA. Under a certain stress, phosphorylation
takes place at the Ser15 site of p53, and promotes transcription of
p21 that eventually causes G1 arrest. In other cases, it promotes activation
of apoptosis inducing genes, such as pig-3, and results in apoptosis. For
those cells that entered G1 arrest, DNA-PK and ATM activity are lost as
soon as DNA is repaired. The loss of DNA-PK and ATMactivity decreases
phosphorylation of p53, so p53 will bind with MDM2 and dissolve.
Without phosphorylation, the p53 protein promotes mdm-2 transcription.
It is interesting to know that mdm-2 protein forms a complex to deac-
20 Hiroaki Kitano
Normal
Protein
Misfolded
Protein rpoH
hsp
Heat
Shock
dnaK
dnaJ
grpE
GroES
GroEL
dnaK
dnaJ
grpE
GroES
GroEL
σ70
σE
σ32
σ32 E
σ32 dnaJ
dnaK
grpE
Figure 1.4 Heat shock response with feedforward and feedback control
tivate the p53 protein. This is another negative feedback loop embedded
in this system.
Redundancy
Redundancy also plays an important role in attaining robustness of the
system, and is critical for coping with accidental damage to components of
the system. For example, the four independent hydraulic control systems
in a Boeing 747 render the systems functionally normally even if one or
two of them are damaged. In aircraft, control systems and engines are
designed to have a high level of redundancy. In a cellular system, signal
transduction and cell cycle are equivalent to control systems and engines.
A typical signal transduction pathway is the MAP kinase cascade.
The MAP kinase cascade involves extensive cross talk among collateral
pathways. Even if one of these pathways is disabled due to mutation or
other causes, the function of the MAP kinase pathway can be maintained
because other pathways still transduce the signal (Figure 1.6).
Cell cycle is the essential process of cellular activity. For example, in
the yeast cell cycle, the Cln and Clb families play a dominant role in the
progress of the cell cycle. They bind with Cdc28 kinase to form Cdk complex.
Cln is redundant because knock-out of up to two of three Cln (Cln1,
Cln2, Cln3) does not affect the cell cycle" all three Cln have to be knocked
out to stop the cell cycle. Six Clb have very similar features, and may have
originated in gene duplication. No single loss-of-function mutant of any
of the six Clb affects growth of the yeast cell. The double mutants of CLB1
21 Systems Biology: Toward System-level Understanding of Biological Systems
p53
DNA-PK
mdm2
p53
MDM2
p53
MDM2
p53
p
pig-3,
etc. Apoptosis
p53
p
p21 G1 arrest
DNA damage
DNA repair
during G1 arrest
ATM
ATM
p
Figure 1.5 p53 related feedback loop
and CLB2, as well as CLB2 and CLB3s are lethal, but other double mutant
combinations do not affect phenotype. It is reasonable that the basic mechanism
of the cell cycle has evolved to be redundant, thus robust against
various perturbations.
Redundancy can be exploited to cope with uncertainty involved in
stochastic processes. McAdams and Arkin argued that duplication of
genes and the existence of homologous genes improve reliability so that
transcription of genes can be carried out even when only a small number
of transcription factors are available (McAdams and Arkin, 1999). The use
of a positive feedback loop to autoregulate a gene to maintain its own expression
level is an effective means of ensuring the trigger is not lost in the
noise.
Although its functional implication has not been sufficiently investigated,
an analysis of MAP kinase cascade revealed that it utilizes nonlinear
properties intrinsic in each step of the cascade and positive feedback
to constitute a stable all-or-none switch (Ferrell and Machleder, 1998).
In the broader sense, the existence of metabolic pathways that can alternatively
function to sustain cellular growth with changing environment
can be viewed as redundancy. Bacteria is known to switch metabolic pathways
if deprived of one type of nutrition, and to use other types of nutrition
that are available. Theoretical analysis combined with experimental
data indicate that different pathways are used to attain essentially the
22 Hiroaki Kitano
Raf, Mos MEKK1, MLK3 ASK1, TAK1
MEK1,2/MKK1,2 SEK1,2/MKK4,7 MKK3,6
MAPK/ERK SAPK/JNK p38
Transcription
Figure 1.6 Redundancy in MAP kinase cascade
same objective function (Edward et al., 2001).
Once we understand the stability and robustness of the system, we
should be able to understand how to control and transform cells. We will
then be ready to address such questions as how to transform cells that are
malfunctioning into normal cells, how to predict disease risk, and how to
preemptively treat potential diseases.
Modular Design
Modular design is a critical aspect of the robustness: it ensures that damage
in one part of the system does not spread to the entire system. It may
also ensure efficient reconfiguration throughout the evolutionary process
to acquire new features.
The cellular structure of the multicellular organism is a clear example.
It physically partitions the structure so that the entire system does not
collapse due to local damage.
Gene regulatory circuits are considered to entail a certain level of modularity.
Even if part of the circuit is disrupted due to mutation or injection
of chemicals, it does not necessary affect other parts of the circuit. For example,
mutation in p53 may destroy the cell cycle check point system that
leads to cancer. However, it does not destroy metabolic pathways, so the
cells continue to proliferate. How and why such modularity is maintained
is not well understood at present.
Modularity reflects hierarchical organization of the system that can be
viewed as follows:
Component: An elementary unit of the system. In electronics, transistors,
capacitors, and resistors are components. In biological systems, genes and
proteins, which are transcriptional products, are components.
Device: A minimum unit of the functional assembly. NAND gates and
23 Systems Biology: Toward System-level Understanding of Biological Systems
flip-flops are examples of devices3. Transcription complexes and replication
complexes are examples of devices. Some signal transduction circuits
may be considered as devices.
Module: A large cluster of devices. CPU, memory, and amplifiers are
modules. In biological systems, organella and gene regulatory circuits for
the cell cycle are examples of modules.
System: A top-level assembly of modules. Depending on the viewpoint,
a cell or entire animal can be considered as a system.
In engineering wisdom, each low-level module should be sufficiently
self-contained and encapsulated so that changes in higher-level structure
do not affect internal dynamics of the lower-level module. Whether is this
also the case for biological systems and how it can be accomplished are of
major interest from a system perspective.
Structural Stability
Some circuits may, after various disturbances to the state of the system,
resume as one of multiple attractors (points or periodic). Often, feedback
loops play a major role in making this possible. However, feedback does
not explicitly control the state of the circuit in tracking or adapting to
stimuli. Rather, dynamics of the circuit exhibit certain functions that are
used in the larger sub-systems.
The most well understood example is seen in one of the simplest
organisms, lambda phage (McAdams and Shapiro, 1995). Lambda phage
exploits the feedback mechanism to stabilize the committed state and
to enable switching of its pathways. When lambda phage infects E. coli,
it chooses one of two pathways: lysogeny and lysis. While a stochastic
process is involved in the early stage of commitment, two positive and
negative feedback loops involving CI and Cro play a critical role in stable
maintenance of the committed decision. In this case, whether to maintain
feedback or not is determined by the amount of activator binding to the
OR region, and the activator itself cuts off feedback if the amount exceeds
a certain level. This is an interesting molecular switch that is not found
elsewhere. Overall, the concentration mechanism of Cro is maintained
at a certain level using positive feedback and negative feedback. It was
reported that the fundamental properties of the lambda phage switch
circuit are not affected even if the sequence of OR binding sites is altered
(Little et al., 1999). This indicates that properties of the lambda phage
decision circuit are intrinsic to the multiple feedback circuit, not specific
parametric features of the elements, such as binding sites.
Relative independence from specific parameters is an important fea-
3 In electronics, “device” means transistors and other materials mentioned in “components.”
NAND gates and flip-flops are recognized as minimum units of the circuit.
24 Hiroaki Kitano
ture of a robust system. Recent computational studies report that circuits
that are robust against a broad range of parameter variations are found
in Xenopus cell cycle (Morohashi et al., unpublished) and body segment
formation (von Dassow et al., 2000). Using the simulation of parasegment
formation of Drosophila, it was found that some parameters in the circuit
accountable for pattern formation are tolerant to major parameter variations.
This strongly suggests that the structure of the circuit that is dominantly
responsible for pattern formation rather than specific parameter
values (von Dassow et al., 2000).
Such circuit features of structural stability also play important roles
in development. A recent review article (Freeman, 2000) elucidates some
interesting cases of feedback circuits that play a dominant role in the development
process. Such cases include temporal arrangement of signaling
in the JAK/STAT signaling pathway, pattern formation in Drosophila
involving Ubx and Dpp, maintenance of patterns of expression for sonic
hedgehog (Shh) that forms ZPA and Fgf, forming AER in limb development,
etc. In these examples, structure of circuits play the dominant role
rather than specific set of parameters.
THE SYSTEOME PROJECT
In order to promote scientific research of systems biology, it is critically
important to create a comprehensive data resource that describes systems’
features, as does the human genome project. This is an enormous
challenge, and it requires significant efforts far beyond the capability of
any single research group. Therefore, the author proposes “The Systeome
Project” as a grand challenge in the area of systems biology.
Systeome is an assembly of system profiles for all genetic variations
and environmental stimuli responses. A system profile comprises a set of
information on the properties of the system that includes the structure of
the system and its behaviors, analysis results such as phase portfolio, and
bifurcation diagrams. The structure of the system includes the structure
of gene and metabolic networks and its associated constants, physical
structures and their properties.
Systeome is different from a simple cascade map, because it assumes
active and dynamic simulations and profiling of various system statuses,
not a