Tài liệu Báo cáo khoa học: Applications and trends in systems biology in biochemistry docx

REVIEW ARTICLE

Applications and trends in systems biology in

biochemistry

Katrin Hu¨ bner, Sven Sahle and Ursula Kummer

Department of Modeling of Biological Processes, COS Heidelberg/BioQuant, University of Heidelberg, Germany

Keywords

metabolism; modeling; quantitative

experiments; signaling; simulation; systems

biology

Correspondence

U. Kummer, Department of Modeling of

Biological Processes, COS Heidelberg/

BioQuant, University of Heidelberg, Im

Neuenheimer Feld 267, 69120 Heidelberg,

Germany

Fax: +49 6221 5451483

E-mail: ursula.kummer@bioquant.

uni-heidelberg.de

(Received 10 January 2011, revised 31 May

2011, accepted 15 June 2011)

doi:10.1111/j.1742-4658.2011.08217.x

Systems biology has received an ever increasing interest during the last

decade. A large amount of third-party funding is spent on this topic, which

involves quantitative experimentation integrated with computational

modeling. Industrial companies are also starting to use this approach more

and more often, especially in pharmaceutical research and biotechnology.

This leads to the question of whether such interest is wisely invested and

whether there are success stories to be told for basic science and/or technol￾ogy/biomedicine. In this review, we focus on the application of systems

biology approaches that have been employed to shed light on both

biochemical functions and previously unknown mechanisms. We point out

which computational and experimental methods are employed most

frequently and which trends in systems biology research can be observed.

Finally, we discuss some problems that we have encountered in publica￾tions in the field.

Introduction

One of the fastest growing fields in the life sciences is

systems biology. PubMed lists more than 3000 articles

which, in one way or the other, use this term in their

title or abstract during the last decade (precisely, the

last 11 years, including the year 2000) compared to a

mere three articles in the preceding century. Obviously,

this is partially a result of the fact that the term ‘sys￾tems biology’ had not been used during that time.

However, as we will see in the present review, also

with respect to research that would now be called sys￾tems biology, there is clearly significantly less to report

before the year 2000. Interestingly, looking closely at

the more than 3000 articles using the term ‘systems

biology’, it becomes apparent that approximately half

of them describe methodological work either on the

computational or the experimental side, and more than

one-third are classified as reviews. However, only a

handful of the latter represent reviews that actually

review a set of articles. Most of the articles classified

as reviews could rather be classified as news and views.

Another large portion of articles uses the term ‘systems

biology’ in a different sense than we would understand

it (e.g. stating that they are investigating a biological

system and it is therefore systems biology). This latter

point necessitates the definition of the term ‘systems

biology’ as we (the authors) understand it, as outlined

below.

Systems biology combines quantitative experimental

data from complex molecular networks (e.g. biochem￾istry, cell biology in the living cell) with computational

modeling. Here, computational modeling does not

refer to statistical models or models of data mining

but rather to a mathematical or ’virtual’ representation

of the living system of interest in the computer, where

Abbreviations

FBA, flux balance analysis; ODE, ordinary differential equation; PDE, partial differential equation..

FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS 2767

there is also a correspondence between parts of the

biological system and parts of the model. This

representation allows a computational analysis using

systems theoretical approaches.

This definition is probably shared by many scientists

in the field [1,2]. The actual term ‘systems biology’ was

coined in 1968 by Mesarovic´ [3]. Soon afterward, the

first conceptional developments on the theoretical side

layed the foundation of the field, such as metabolic

control analysis [4,5] and biochemical systems theory

[6]. In the 1980s, the development of extreme currents

and elementary modes [7,8] and stochastic frameworks

[9] followed. These conceptional approaches were then

implemented in specialized software tools, as will be

seen below.

However, to identify articles encompassing applica￾tions of systems biology approaches that fit this defini￾tion, we note that, on the one hand, it is completely

insufficient to search for articles that explicitely state the

term ‘systems biology’. On the other hand, it is extre￾mely difficult to define good keywords for a search in

PubMed because the term ‘model’, as well as similar

terms, are used in many different contexts and it is very

cumbersome to find relevant work in the multitude of

articles that are available with obvious keywords.

Therefore, we first defined the scope of the articles

that we would like to review. These have to fit the

above definition in the sense that they represent exam￾ple cases of applying systems biology approaches com￾bining experimental investigation and computational

modeling (subsequent to the year 2000). In addition,

fitting our own expertise and the scope of the FEBS

Journal, we restrict ourselves to typical intracellular

biochemical systems. These include signaling systems

and metabolic pathways. Here, models have to

describe explicit biochemical mechanisms of systems

and have to relate to quantitative experimental mea￾surements of systems behaviour appearing in the same

article or in previous publications. Correspondingly,

purely experimental findings have to directly relate to

previous computational models.

We do not focus on cell biological, biomechanical or

higher level descriptions of multicellular systems in the

present review. Finally, the systems biology of the cell

cycle and of circadian rhythms have been properly

reviewed recently [10,11] and therefore we do not

include them here. With this scope in mind, we opti￾mized a keyword search for PubMed with the following

limits: year AND [in silico OR biology OR biochem*

OR bioinformatic* OR biological OR intracellular OR

biophysic* AND (modeling OR modeling OR ‘mathe￾matical model’ OR ‘mathematical models’ OR ‘kinetic

model’ OR ‘kinetic models’ OR ‘differential equation

model’ OR ‘multiscale model’ OR ‘dynamic model’ OR

‘quantitative model’ OR ‘computational model’ OR ‘petri

net model’ OR ‘agent based model’ OR ‘stochastic

model’ OR ‘flux balance’ OR ‘dynamical model’ OR

‘homeostatic model’ OR (model AND simulation*)]

NOT ‘protein structure’ NOT ‘animal model’ NOT

review[publication type] AND (metabolism OR meta￾bolic OR signal* OR ‘cell cycle’ OR oscillation*) NOT

pharmacokinetic* NOT pharmacodynamic* NOT elec￾trophysiolog* NOT ‘molecular modeling’ NOT ‘molecu￾lar modeling’ NOT ‘homology modeling’ NOT

‘homology modeling’ NOT ‘MD simulation’ NOT

‘molecular dynamics’).

This search resulted in approximately 17 000 articles

of which we read the titles and abstracts and, in cases

of doubt, the article as such to select the relevant ones,

resulting in the approximately 400 articles that we

review. Even though we try to be as complete as possi￾ble, it is obvious that we employed heuristics with the

above strategy and also certainly and unintentionally

missed one or more articles. However, checking

against, for example, the BioModels database [12],

which contains a curated collection of biological mod￾els, and against older reviews that review the field par￾tially and from a different viewpoint [13–16], we

estimate that we cover at least a representative

80–90% of those articles in the field that fit the above

requirements. Thus, we offer a good picture of the

field with respect to the last decade.

Similar to the highly informative review about math￾ematical modeling of metabolism by Gombert and

Nielson [17], all articles are summarized extensively in

tabular form to allow a quick overview of the pub￾lished material. Table 1 provides information on the

studied system, major findings, and employed compu￾tational and experimental approaches, as well as the

reference itself. Figure 1 provides a tree-like view on

how the articles are ordered to ease navigation within

Table 1 itself. The ordering is by systems because

many scientists will be interested in a specific system,

even across species boundaries. The large number of

articles reviewed prohibits a detailed referencing in the

text when discussing general trends. For recapitulating

these trends, we would make reference to Table 1.

General developments

There is a clear increase in publications that employ

systems biology approaches to tackle open biochemical

questions. Because we focused on original work, rather

than on any articles just mentioning systems biology,

this fact is not blurred by the vastly increasing number

of news and views, articles and minireviews, and so

Systems biology in biochemical research K. Hu¨bner et al.

2768 FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS

on. The number of articles appearing annually within

the last few years is approximately four-fold greater

than in the year 2000 (Fig. 2). Before 2000, there are

only few articles that actually would fall into the above

category, as quickly checked by the same query. Of

course, many valuable modeling articles had been pub￾lished before 2000, although very few of these worked

directly with quantitative biological data. One of the

exceptions is the field of calcium signaling, where com￾putational modeling very quickly formed the basis for

deciphering the mechanism behind calcium oscillations

[18].

In addition to the general trend to use systems biol￾ogy approaches more frequently, there is also an

increasing trend in the articles to actually validate the

developed models with experimental data. This is defi￾nitely a positive development because the actual vali￾dation of the computational models aids in an

assessment of their reliability.

The number of journals publishing systems biology

work is also increasing, although there are only a few

journals that often appear in our data. The most

me

Fig. 1. Systematic tree for the navigation of Table 1. Articles are ordered according to the system studied and systems are annotated with

gene ontology (GO) numbers.

0

10

20

30

40

50

60

2000

2001

2002

2003

2004

2005

2006

2007

2008

2009

2010

# publications

Year

Articles

Fig. 2. Number of publications describing systems biology applica￾tions in biochemistry per year.

K. Hu¨bner et al. Systems biology in biochemical research

FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS 2769

common ones covering the whole period (Fig. 3) are

Biophysical Journal, Journal of Theoretical Biology, Bio￾technology and Bioengineering, FEBS Journal (formerly

European Journal of Biochemistry), Journal of Biological

Chemistry and Metabolic Engineering. Within the last

few years, more specialized journals have established

themselves. Here, the most frequently appearing ones

are BMC Systems Biology, Molecular Systems Biology

and PLoS Computational Biology. There is a clear trend

from the more engineering-oriented journals to the basic

research-oriented ones over the years.

Often, systems biology articles are quite long, which

is a result of the fact that they have to describe both

experimental and computational methodology, as well

as the results from both. Similar to many other fields,

this has led to a rather annoying trend, namely putting

extensive material into a supplement. This results in

articles that are almost uncomprehensible without

reading the supplementary material as well. Very often,

the actual model that is the basis for the results, and

thus is an absolutely crucial part of the work, ends up

in the supplementary information. Even though it is

often possible to download this material along with

the original article, it does not make the reading of a

scientific work any easier by pushing central informa￾tion into an additional file. The least that journals

should consider is an automated packaging of both

files into one pdf for download. Fortunately, this has

already been implemented for least a few journals (e.g.

Nature, Journal of Biological Chemistry). One addi￾tional issue arising with this policy is the fact that

references cited in the supplementary material do not

count for citation indices and the computation of

h-indices, etc. The latter was confirmed by us by

testing different examples from several journals. Plac￾ing formulations of models as well as crucial method￾ology, both on the experimental and computational

sides, into the supplementary material then implies a

strong and systematic disadvantage for the careers of

young scientists working in these fields.

Systems studied

The organisms studied with systems biology

approaches in the last decade are by a large extend

eukaryotic and only to a lesser extent prokaryotic

(Fig. 4). Among the first, classical scientific model

organisms such as Saccharomyces cerevisiae, Mus mus￾culus, Rattus norvegicus and, for obvious reasons,

Homo sapiensare dominant. However, studies also

include the parasite Trypanosoma brucei [19,20] or the

biotechnologically relevant Aspergillus niger [21–24].

Again, the prokaryotic key players are typical model

organisms, such as Eschericia coli, although biotechno￾logically relevant organisms, such as Lactococcus lactis

and Corynebacterium glutamicum, are often investi￾gated. Prokaryotic organisms of medical relevance,

such as Mycobacterium tuberculosis [25,26] and Heliob￾acter pylori [27,28] appear twice, with many others

only appearing once.

The biochemical networks that are studied in these

prokaryotic organisms have been mostly of metabolic

0

5

10

15

20

25

30

35

Biophysical journal

BMC systems biology

Molecular systems biology

Journal of theoretical biology

PLoS computational biology

Biotechnology and bioengineering

FEBS journal

Journal of biological chemistry

Metabolic engineering

PLoS one

# publications

Journals

Fig. 3. Number of publications describing

systems biology applications in biochemistry

in the years 2000–2010 in the 10 most

often used journals.

Systems biology in biochemical research K. Hu¨bner et al.

2770 FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS

nature, reflecting their importance in biotechnology.

Here, apart from the central energy metabolism includ￾ing glycolysis (Fig. 5), pathways of biotechnological

importance such as lysine synthesis [29] in Corynebac￾terium glutamicum, sucrose synthesis [30–32] in sugar

cane, xanthan biosynthesis in Xanthomonas campestris

[33] and citrate metabolism in fruit [34] have been

studied.

By contrast, most studies on eukaryotic (e.g. mam￾malian and especially human) cells focus on signaling

systems, which reflects the importance of these systems

in the context of cancer research. Dominant examples

are calcium, nuclear factor jB, extracellular signal-reg￾ulated kinase, mitogen-activated protein kinase and

janus kinase-signal transducer and activator of tran￾scription signaling (Fig. 5).

There is a clear trend towards eukaryotic and signal￾ing systems over the years, which coincides with

the above observation that basic medical science has

discovered systems biology later than the engineering

field, in which metabolic engineering has been one of

the forerunners. Signaling pathways are either studied

in isolation or, with increasing numbers, in an integra￾tive way, encompassing several pathways and their

cross-talk. Unexpectedly, only few articles feature a

combination of signaling and metabolic networks.

However, these are also increasing slowly.

Thus, the overall picture depicts more specific meta￾bolic systems studied in the beginning of the decade,

often published in biotechnology/engineering journals.

Later, signaling systems became slighty prevalent,

reflecting systems of medical relevance in eukaryotic

cells. Finally, with the whole genome-based metabolic

models becoming more approachable from approxi￾mately 2005 onwards, metabolism has been catching

up again (Fig. 6).

Experimental approaches

Here, we focus on the experimental approaches used in

conjecture with computational modeling, in the core of

a systems biology approach.

Experimental data in systems biology are obviously

either time-series data (if used for dynamic models) or

single time point data (if used for static models). How-

0

10

20

30

40

50

60

70

Genome-scale

Central

Carbohydrate

Energy

Amino acid

Calcium

I-κB/NF-κB

MAPK (ERK)

JAK-STAT

Apoptosis

# publications

Metabolism

Signaling

Fig. 5. Number of publications describing

systems biology applied to specific bio￾chemical systems in the years 2000–2010.

0

10

20

30

40

50

60

70

80

90

H. sapiens

S. cerevisiae

M. musculus

E. coli

R. norvegicus

L. lactis

C. glutamicum

X. laevis

A. niger

A. thaliana

# publications

Organisms

Fig. 4. Number of publications describing systems biology applied

to the study of specific organisms in biochemistry in the years

2000–2010.

K. Hu¨bner et al. Systems biology in biochemical research

FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS 2771

ever, in some cases, dynamic models are also build

using steady-state profiles. This is true for data used as

a basis for modeling, as well as for data used for

model validation.

The compounds commonly measured in time-series

analysis are metabolites (hereon, we refer to all chemi￾cal species other than macromolecules as metabolites),

proteins and, to a lesser extent (in the light of the pres￾ent reviewed systems), RNA and DNA. In addition,

enzymatic activities and cellular properties such as

growth and death rates are measured in a time-depen￾dent manner.

Only a very few metabolites are measured in vivo

(e.g. using imaging technologies). Examples that fre￾quently are measured using in vivo methods are cal￾cium (in the more than 30 publications studying

calcium signaling) and NADPH [35]. In only a few

cases, NMR is also employed for in vivo studies [36–

39]. However, most often, metabolites are extracted

from cells and measured in vitro. This puts limits on

the time resolution of the experimental results, which

does not allow fast dynamics to be followed. In many

cases, the temporal dynamics of the system of studied

is rich over a relatively short time-scale (e.g. calcium,

p53, NF-jB, nuclear factor jB), which was only dis￾covered after in vivo methods became available for

these compounds. Together with the relatively high

level of noise in many of the in vitro measurements,

this highlights the need for a strong effort to develop

new methods for detecting metabolites in vivo, such as

the development of nanosensors [40], with the expecta￾tion that many as yet unknown behaviours will be

discovered subsequently.

The in vitro characterization of metabolites after pre￾paring cell extracts is mostly carried out using HPLC

or assay kits and, in a few cases, with GC-MS.

The dominant technology to measure protein con￾centrations is immunoblotting. Approximately 70% of

all manuscripts featuring protein concentrations (e.g.

in the context of signaling) use this method, which

again requires cells to be killed and their contents

extracted. Therefore, it is quite unexpected that live

cell imaging methods for proteins (e.g. using green

fluorescent protein-tagged antibodies) are also still only

rarely used in systems biology studies.

Obviously, live cell imaging on the one hand is also

hampered by several problems (e.g. the need to follow

many cells to be able to judge cell–cell variation, signal

to noise ratios with proteins or metabolites of low con￾centrations and the autofluorescence of some cell

types). On the other hand, in vitro measurements are

limited by the above mentioned facts, such as low time

resolution and experimental errors and, in addition,

these methods are often so laborous and expensive

that they are only performed in up to three replicas

with computed standard deviations that have dubious

statistical meaning. Often, replicas are purely technical

and not biological replicas.

Enzyme activities are usually measured with

standard kits. If these are measured in cell extracts or

in vitro under physiological conditions, they are a

valuable source for the modeling. However, studies

0

10

20

30

40

50

60

2000 2001 2002 2003 2004

2005 2006

2007 2008 2009

2010

2000 2001 2002 2003 2004

2005

2006 2007

2008 2009 2010

# publications

Metabolism

Genome-scale metabolism

Signaling

Metabolism + signaling

Prokaryotes Eukaryotes

Fig. 6. Number of publications per year

describing signaling, metabolic systems,

whole-genome metabolic models or mixed

systems in prokaryotic and eukaryotic

organisms, respectively.

Systems biology in biochemical research K. Hu¨bner et al.

2772 FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS

frequently refer to kinetic parameters measured in test

tubes using isolated enzymes under highly unphysio￾logical conditions as the basis for an initial parameter

guess, although these often have been shown to be far

away from actual in vivo parameters [41].

Computational approaches

Studying the computational approaches used in the

systems biology of cellular biochemistry, it is highly

obvious that the formalism of ordinary differential

equations (ODE) is the dominating approach (Fig. 7).

This does not necessarily mean that the scientist

actually set up ODEs by him/herself because several

software tools used in systems biology allow a process￾based modeling (e.g. the entry of a reaction scheme)

and translate this reaction scheme into ODEs. How￾ever, temporal or dynamic models are mainly simu￾lated and analyzed in this mathematical framework.

All other approaches do not yet play a significant role.

Nevertheless, stochastic approaches are specifically

used in the context of signaling networks because these

networks often feature low copy numbers of molecules,

which poses problems for the ODE framework. Static

or stoichiometric models are mainly analyzed using

flux balance analysis (FBA), which has become the sec￾ond most abundant computational approach in recent

years.

Unexpectedly, few models describe spatial as well as

temporal developments of biochemical systems. This

might be the result of a variety of factors: First, corre￾sponding experimental data are still sacrve. Second,

computational methods (e.g. for the parametrization of

the models) are much less developed than for ODE

based models. Furthermore, there are fewer user￾friendly software tools that allow spatial modeling

and, thus, more programming is required for this type

of modeling. This is also reflected by the fact that no

increase in the usage of spatial models has been

observed over the last 10 years. Unless more user￾friendly tools become available, we consider that there

will be no clear trend in this direction. For the few

spatial models available, the dominating computational

approach is the use of partial differential equations

(PDEs).

The computational tasks applied on the temporal or

dynamic models are mostly simulations, the fitting of

model parameters to experimental data and the

computation of sensitivities to detect dependencies in

the model. Here, parameter estimation is rarely and only

recently linked to a discussion of parameter

identifiability, which appears to enter the field only now.

This certainly should have more impact in the future.

Very often, the exact methodology by which these

computations are carried out is not documented in the

articles. We find it utterly unexpected that, overall, it

is only a minority of articles that properly describe

(in a reproducible way) the computational research

performed in the study. Thus, very often, neither the

exact numerical algorithm used to simulate a specific

behaviour, nor the software with which the computa￾tion was performed, are given and referenced. This has

somewhat improved over the course of the decade,

although it appears that there is a lack of awareness of

the fact that a documentation of the computational

approaches is scientifically as important as the docu￾mentation of the experimental data, which are never

missing. This problem is increased by the trend (as

noted above) of some journals to put crucial (e.g.

methodological) information, and sometimes even the

whole description of the computational model, into the

supplementary material. Once again, this renders arti￾cles incomprehensible without reading the supplement

and puts those scientists who are working on new

methods and tools into the unfortunate situation that

their work might only be cited in the supplement,

which does not appear in the science citation index.

Accordingly, it is very hard to review the trends within

the algorithms and tools. It is, however, clear that the

commercial software matlab (MathWorks, Natick,

MA, USA; www.mathworks.com) is the dominating

software (Fig. 8). Additional commercial software

packages that are widely used are mathematica (Wol￾fram Research, Champaign, IL, USA; www.wolfram.-

com) and, for the set-up and analysis of whole-genome

0

50

100

150

200

250

ODE

Stoichiometric

PDE

Stochastic

Logic

Petri net

Hybrid

# publications

Modeling methodology

Fig. 7. Number of publications describing systems biology applied

to biochemistry in the years 2000–2010 using a specific computa￾tional modeling approach.

K. Hu¨bner et al. Systems biology in biochemical research

FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS 2773

models, lindo (Lindo Systems Inc., Chicago, IL,

USA; www.lindo.com) and simpheny (genomatica,

San Diego, CA, USA; www.genomatica.com). In addi￾tion, free and specialized software, such as xppaut

[42], copasi [43] and gepasi [44], as well as the semi￾academic software berkeley madonna [45], are being

used more and more often.

The above observation about poorly documented

computational methodology unfortunately also applies

to models themselves. Thus, often important parame￾ters (e.g. initial values) are missing and sometimes

incomplete equations are given. Here, it should be

mentioned that a very few journals (e.g. FEBS Journal)

actually employ curation of models submitted for pub￾lication via usage of JWS Online [46], which helps to

avoid these problems.

Two trends within the last few years are positive

and interesting. First, slowly, more and more models

receive proper validation within the study. This means

that the model is not only used to reproduce data

(often after parameter fitting), but also is actually used

for independent predictions of observable behaviour,

which is then experimentally verified and thus the

model is validated. The second trend is the re-use of

models. Thus, more and more studies rely on previous

modeling work, either by extending or modifying exist￾ing models, or by merging existing models with each

other or with new models. This trend is supported by

and necessitates the development of software standards

for the exchange (sbml [47], cellml [48]) and docu￾mentation of models (miriam [49], as well as central

data resources for the storage of computational

models, such as the well curated BioModels database

[12], JWS Online [46], the CellML repository [50] or,

for whole-genome scale models, the BIGG database

[51]). These approaches will hopefully help to over￾come problems of insufficient documentation, at least

on the model side. On the side of computational meth￾ods, there is currently a similar community effort that

creates a standard for minimal information called

MIASE [52].

Finally, we would like to mention that by and large

our results agree with an analysis of currently used

computational standards, approaches and tools that

was based on questionaires distributed to computa￾tional scientists in the field and published in 2007 [53].

However, because of the differring nature of data gen￾eration, there are also a few significant differences (e.g.

approaches) that are rarely mentioned in published

research (as in the present review) and are more often

named in the questionaires. As an example, probabilis￾tic approaches occur at least in 20% of the questio￾naire responses, although they are significantly less

prevalent in the publications reviewed here. A similar

situation applies to some software tools that are more

dominant in the questionaire-based survey and are

scarcely noted in the actual publications.

Discussion

The last decade has seen a strong increase in research

carrying the label systems biology, which combines

computational and quantitative experimental investiga￾tions at a systems level. On the one hand, we were sur￾prised by the fact that only a small fraction of the

publications found using the keyword ’systems biology’

actually reflect applications of systems biology

approaches to biological systems resulting in new bio￾logical insights. However, on the other hand, and by

restricting ourselves to purely biochemical applications,

we identified almost 400 publications that represent

successful applications of systems biology, and the

numbers are clearly on the rise. The success of these

applications is obviously often visible as a scientific suc￾cess and only rarely as a success that results directly in

biotechnological or pharmaceutical developments.

However, this is of course true for most scientific disci￾plines. Stating that these are successful applications

does not imply that all of the cited articles are very

strong cases; many are and some are not.

However, our aim is to give a comprehensive and

representative overview of systems biology research, its

trends and the commonly used computational, as well

as experimental, methodologies. Therefore, we decided

not to focus on just a few articles but, rather, to try to

gather a complete as possible set of publications.

0

20

40

60

80

100

120

Matlab

NG

SimPheny

LINDO

COPASI

XPPAUT

Mathematica

Gepasi

Own

COBRA

Berkeley

madonna

# publications

Software

Fig. 8. Number of publications describing systems biology applied

to biochemistry in the years 2000–2010 employing the ten most

commonly used software tools.

Systems biology in biochemical research K. Hu¨bner et al.

2774 FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS

When compiling this review, we came across a num￾ber of unexpected problems, some of which we have

already noted above. Missing documentation of com￾putational research is a clear and abundant problem

that makes systems biology research less tractable than

it should be. In our opinion, this must change. In addi￾tion, terminologies in such an interdisciplinary field

have to be chosen with care. To exemplify this point,

in many publications, the term ‘experiment’ is used for

a computational experiment (e.g. a simulation). This is

quite normal in theoretical or mathematical literature.

However, in the context of systems biology, this is con￾fusing because it is sometimes not so easy to judge, if the

word experiment’, without reference to computations

(e.g. not using the more explicit term ‘computational

experiment’), actually refers to wet-laboratory or dry￾laboratory experiments. Therefore, articles should

either clearly emply the term ‘computational experi￾ments’ when refering to these or use the more com￾monly used terminology (e.g. ‘simulations’). Another

confusing term is ‘prediction’ because some articles use

this word to indicate that their model fits experimental

data (after parameter fitting), whereas, usually, the

term is needed to state that the model actually predicts

experimental behaviour to which it has not been fitted

in the first place. It is sometimes almost impossible to

tell the difference, if it is not clearly indicated which

data have been used for fitting and which have been

used for model validation.

We would like to pick up a question raised at the

beginning of this review: does systems biology repre￾sent an approach that offers anything beyond the

existing purely experimental approaches? Reading the

approximately 400 articles featured in this review, we

would answer with a clear ’yes’. This does not mean

that all studies published have gained many new

insights from the integration of computational model￾ing with quantitative experimentation, although the

majority clearly do. In many studies, computational

modeling is used to understand complex mechanisms

that are difficult to dissect by pure experimental means

and to generate likely hypotheses that push forward

our comprehension of the complicated interactions and

their functionality in quite an efficient way. There are

many examples for this and we only want to highlight

a few of them. One of the prominent examples is the

field of calcium signal transduction where our current

understanding of the mechanism behind the often

observed calcium oscillations would not have been

possible without computational modeling, with this

having already started way before the onset of systems

biology, as reviewed here. However, important new

insights have been generated in the past decade. Thus,

the impact of calcium dynamics on CaMKII has been

studied in detail (see entry 210 in Table 1). Other

downstream effects have been investigated, including

apoptosis (see entry 229 in Table 1). In addition, the

stochasticity of single calcium channels and its impact

on the overall dynamics have been investigated in

many studies (see entry 314 in Table 1).

Further signal transduction systems that exhibit

complex behaviour have been explained quite well with

the aid of validated computational modeling. We are

only able to mention a few examples and, once again,

have to refer to the material in Table 1. A beautiful

study explains the response of yeast to osmotic shock

(see entry 382 in Table 1). The control of MAPK sig￾naling has also been predicted and experimentally con￾firmed (see entry 334 in Table 1). Recently, receptor

properties that are crucial for the information process￾ing within erythropoietin signaling are also identified

(see entry 259 in Table 1).

On the metabolic side, exciting examples of integrated

systems biology approaches are the identification of key

players in the branched amino acid metabolism in Ara￾bidopsis thaliana (see entry 3 in Table 1), understanding

the metabolism of tobacco grown on media containing

different cytokines (see entry 176 in Table 1) and the

investigation of substrate channeling in the urea cycle

(see entry 191 in Table 1).

However, and apart from this more basic scientific

benefit, namely the increased understanding of

complex mechanisms, there are also very applied

examples of research benefitting from systems biology.

Thus, systems biology has been used for the prediction

of drug targets (e.g. see entries 84, 104 and 197 in

Table 1) and for biotechnological engineering (e.g. see

entries 14, 16, 36 and 392 in Table 1). Obviously, most

of these have not entered industrial production yet

(more time is needed for that) but it is clear that sys￾tems biology has become a tool for enabling the dis￾coverery of new potential applications, similar to

molecular modeling and bioinformatics in the past.

Finally, we want to stress once more that we have

restricted ourselves to biochemical systems and

excluded systems of cell cycle and circadian rhythms

because these have been reviewed recently [10,11].

Therefore, the actual number of successful systems

biology studies will be several times the amount

reviewed here.

Acknowledgements

We would like to acknowledge the Klaus Tschira

Foundation and the BMBF (Virtual Liver Network

and SysMO) for funding.

K. Hu¨bner et al. Systems biology in biochemical research

FEBS Journal 278 (2011) 2767–2857 ª 2011 The Authors Journal compilation ª 2011 FEBS 2775