All systems go: launching cell simulation fueled by integrated experimental biology data Masanori Arita1,2, Martin Robert1 and Masaru Tomita1 Biological simulation serves to unify the basic elements of systems biology, namely, model selection, experimentation and model refinement. To select biochemical models for simulation, metabolome analysis can be performed using capillary electrophoresis or liquid chromatography coupled with mass spectrometry. In this manner, selected models can be elaborated with temporal/spatial gene and protein expression data obtained from model organisms such as Escherichia coli. The E. coli single gene deletion mutant library (KO collection) and His-tag/GFP-fusion single open reading frame clone expression library (ASKA) are powerful resources for this task. The integration of parallel experimental datasets into dynamic simulation tools forms the remaining challenge for the systematic analysis and elucidation of biological networks and holds promise for biotechnological applications. Addresses 1 Institute for Advanced Biosciences, Keio University, 403-1 Nipponkoku, Daihoji, Tsuruoka, 997-0017 Yamagata, Japan 2 Department of Computational Biology, Graduate School of Frontier Sciences, University of Tokyo, Kashiwanoha 5-1-5 CB05, 277-8561 Kashiwa, Japan Corresponding author: Tomita, Masaru ([email protected])

Current Opinion in Biotechnology 2005, 16:344–349 This review comes from a themed issue on Systems biology Edited by Hans V Westerhoff Available online 22nd April 2005 0958-1669/$ – see front matter # 2005 Elsevier Ltd. All rights reserved. DOI 10.1016/j.copbio.2005.04.004

Introduction Computational models have crucial roles in the analysis of complex biological systems [1,2]. A single type of largescale experimental analysis of molecular interactions or cellular states (e.g. microarray, yeast two-hybrid assay or metabolic profiling) is insufficient for elucidating biological functions in specific biological contexts. Multiple, often heterogeneous sets of experimental data must be integrated and comprehensively analyzed to untangle the complexity of signaling, regulatory and metabolic pathways, together with their crosstalk [3,4]. Toward this goal, systems biology advocates the iterative process of making models, obtaining experimental data, and reconciling discrepancies between model predictions and experimental outcomes. For such comprehensive studies, target Current Opinion in Biotechnology 2005, 16:344–349

species are often bacteria with known genome sequences and a small number of genes [5]. In systems biology, computational analysis is expected to contribute in three ways. First, is the manifestation of each pathway’s role within the context of others. The declarative representation of biological knowledge, in a visual or graphical format, enables us to view the model and has a major role in systems data analysis [6]. Second, computational analysis is used to estimate model components or to assign parameters that are experimentally indeterminable. Abstract, high-level models are refined to more physicochemical, low-level models through this step [7]. Third, computer simulations can predict a system’s behavior under perturbation. Simulation results are used to generate new testable hypotheses and to improve models, and thus close the iterative cycle of systems biology (Figure 1). Computer simulation of biological systems, in a broad sense, should act as the interface to the above three tasks — forming a bridge between computer scientists and experimental biologists and between our understanding and the real-life form (Figure 2). In this review, we introduce recent advances and efforts in both software and experimental platforms that have helped us move toward this integrative goal and, as a case study, introduce our activities in collecting the necessary, comprehensive data.

Metabolome analysis to reveal decomposable models The first step in systematic simulation is the selection of an appropriate model among all possible candidates. This process can be called model search, and is already within the territory of mathematical modeling [8,9]. Usually, metabolic or signaling pathways for simulation are selected on the basis of functional modules. Therefore, modularity or decomposability is an indispensable prerequisite of simulation models. Formally, a model is considered to be decomposable if its interactions in each module are independent of those in others, and each module contains no context-dependent internal states. In other words, each module has well-defined inputs and outputs and behaves deterministically depending on its inputs, as does a logic gate in electronic circuits [10]. The requirement for decomposability is natural, because a simulation model itself corresponds to a subset of the whole cellular process map. The ability to decompose a manifested model must be statically verified before analyzing its network dynamics. For example, whether www.sciencedirect.com

Cell simulation Arita, Robert and Tomita 345

Figure 1

Metabolome analysis Transcriptome/proteome analysis Growth phenotypes etc. CE/LC-MS

Network Metabolic maps Databases Reconstruction model

Data integration Model refinement

Network dynamics Simulation

Interpretation and feedback Current Opinion in Biotechnology

Basic outline of the systems biology approach and simulation. CE/LC-MS, capillary electrophoresis/liquid chromatography mass spectrometry.

certain metabolites can be synthesized in a selected model is one such criterion to be verified at the manifestation level. A straightforward way to identify biochemical modules for modeling is through the measurement of related metabolite concentrations. As most biosynthetic/degradation pathways are well known and conserved across species, we can estimate corresponding modules from the list of regulated metabolites. A comprehensive study of metabolites in this manner is termed metabolome analysis. In metabolome analysis, separation methods such as capillary electrophoresis (CE), liquid chromatography (LC), and gas chromatography have been used in combination with mass spectrometry (MS) to monitor and quantify metabolites present in a sample. As a large number of metabolites are highly polar and charged, standard LC-MS based methods are limited in their

capacity to separate a broad range of metabolites. To overcome this, Soga and co-workers [11–14] developed a series of CE-MS-based methods that allow the selective and quantitative analysis of polar metabolites in complex samples. The methods allowed the detection of close to 1700 different peaks and highlighted quantitative changes in multiple metabolites during Bacillus subtilis sporulation [15]. A similar study is currently ongoing to quantify metabolic changes in Escherichia coli during different growth phases and on different media (T Soga, personal communication). If necessary, further improvement in the separation resolution can be obtained by using a combination of reverse-phase LC followed by CE [16]. For a rapid and comprehensive survey of metabolites, Fourier transform ion cyclotron MS (FTIC-MS) and time-of-flight (TOF) MS infusion-based methods have drawn interest [17–20]. Although the very high mass resolution of FTIC-MS allows a large number of meta-

Figure 2

Experimental biologists

Deterministic/ stochastic modeling

Simulation software Computational analysis

Large-scale constraint-based approach

Visualization Representation

Current Opinion in Biotechnology

A role for computational analysis and simulation software in biology. Simulation software can act as the interface between biologists and biological data representation or static and dynamic mathematical models of biological systems. www.sciencedirect.com

Current Opinion in Biotechnology 2005, 16:344–349

346 Systems biology

bolites to be simultaneously measured, the commonly used infusion MS scheme without in-line separation makes it difficult to quantify metabolites owing to matrix effects or ion suppression. TOF-MS systems, which can be easily linked to a CE system and which have mass resolution approaching that of FTIC-MS, will prove more powerful for the identification and quantification of metabolites. Exhaustive metabolite profiling is still in its infancy; however, one interesting characteristic of the metabolome is that although overall its functional components are highly diverse, they are completely identical across species. In contrast to protein and DNA sequences which vary in composition during the course of evolution, metabolites are structurally identical regardless of the cell type or species. Thus, a suitable method for identifying particular metabolites can be used across any species with little additional cost. Once a complete catalog of metabolites is established and suitable analytical methods are developed (e.g CE-MS/LC-MS etc.) it is simply a matter of surveying their presence or absence in a particular sample.

Data acquisition is our imminent task To complement the static analysis of a biochemical network, microarray-based gene expression data can be used to observe its responses under temporal conditions. Assignment of gene expression data to a known network or module leads to the refinement of network models or to the observation of regulatory and physiological shifts in response to different stimuli [15,21,22,23,24]. For more in-depth analysis of temporal behavior, model components need to be instantiated by assigning appropriate mathematical frameworks. Typically, a low-level description of molecular interactions is based on chemical kinetics and ordinary differential equations (ODEs), such as the classical Michaelis–Menten’s formulation, generalized mass action, or the S-system (Savageau’s Biochemical Systems Analysis). Recent advances in experimental technologies start to provide data for such numerical modeling. As for protein localization and abundance, two exhaustive studies were reported by collaborating groups making use of a conceptually similar endogenous open reading frame (ORF) tagging system in yeast [25,26]. These ingenious, yet simple, experimental systems uncovered the intracellular localization and the absolute quantity of most yeast proteins. Currently available data represent the cellular situation under only a limited number of conditions, and the amount of information is still far less than required for partial differential equation (PDE) models, which consider spatial distributions. Although some simulators support general PDE solvers [27], the lack of appropriate representation of subcellular architectures, and the lack of Current Opinion in Biotechnology 2005, 16:344–349

experimental techniques for measuring parameters in natural, compartmentalized conditions, make PDE simulations far more difficult than their ODE counterparts (see Update). There are many challenges that we have yet to overcome in large-scale data collection and interpretation. Gene expression levels and protein abundance were recently demonstrated to be poorly correlated [28,29], indicating the importance of obtaining data at more than one level of biological information to make sense of the information. The same is probably true for the correlation of intracellular enzyme levels with metabolite levels. Moreover, a recent analysis of yeast functional genomics data accumulated so far has demonstrated how careful we have to be in interpreting sometimes poorly overlapping largescale datasets and the inherent bias of experimental approaches [30]. Another important issue is that most large-scale experiments to date give an idea about the average value of gene, protein or metabolite expression levels in a cell population; for example, it is unknown whether intermediate gene expression levels result from a mixture of individual cells in either the on or the off state [31–33]. Ultimately, what we need is technology for the noninvasive measurement of rates and concentration of reactants in a single cell [34]. Without such measurements, the biological context where interaction data has meaningful roles will not be understood.

Escherichia coli as a model organism for simulations Yeast, one of the simplest eukaryotes, has been the forerunner in many large-scale studies [35]. However, major experimental efforts within the E. coli research community promise to bring E. coli back to the frontline of ‘big’ biology [36] (see Update). As examples, the KO collection — a vast collection of in-frame single gene deletion mutants (3972 genes) of the K-12 strain — is nearing completion (H Mori, personal communication) and is already an exceptional experimental resource for functional genomics and systems biology (http://ecoli. aist-nara.ac.jp/gb5/Resource_download.html). Another similar mutagenesis resource (2008 genes) that is not yet complete uses a transposon-mediated gene inactivation method [37]. In addition, ‘A complete Set of E. coli K-12 ORF Archive’ (ASKA) library is a full-length histidinetagged-ORF clone library with or without green fluorescent protein (GFP) fusion, which can be used to express and purify most E. coli proteins for functional analysis. It has recently been used in functional assays to identify novel nucleotidase activities using 300 different uncharacterized proteins and general enzyme assays [38] and for localizing close to 4000 E. coli proteins using the GFP fusion proteins (http://ecoli.aist-nara.ac.jp/ GB5/search.jsp). Together, we can expect that these www.sciencedirect.com

Cell simulation Arita, Robert and Tomita 347

resources will provide valuable data on an unprecedented scale. To solve larger dynamic puzzles, experimental assays will need to be performed systematically under various conditions. This will, in turn, produce a massive amount of data that needs to be integrated with all the associated challenges [39]. We can imagine a new level of difficulty; experimental procedures that are massively parallel in nature (microarray, proteome analysis etc.) will need to be performed under a variety of starting conditions [4]. Despite this complexity, there are good reasons to believe this goal is achievable, especially for bacteria with smaller genomes [5]. Only when such conditions are fulfilled can we expect to achieve a truly dynamic picture of biological events [39].

From static to dynamic simulation with large-scale data The ability to infer unknown gene functions or to obtain undetermined model parameters depends on the extent to which we restrict the expressive power of models in advance. If we assume the general, nonlinear ODE as a model for genetic interaction, the estimation of parameters from gene expression time-courses already becomes intractable in a system with as few as 20 genes. As a linear, static system, on the other hand, optimization of the entire E. coli metabolic flux is feasible. Considering the number of components to be optimized, the static, constraintbased approach is a good primer for parameter estimation. The validation of a linear model was recently demonstrated by showing that single gene mutants of E. coli evolved in vitro on various substrates to growth phenotypes that could be predicted rather accurately [40,41]. The same authors also showed how apparently disparate datasets of gene expression and metabolic data can be used to generate new testable hypotheses [42]. Such linear models only represent equilibrated, non-dynamic states. What we seek is how to elaborate them into dynamic models with available data. One solution is the use of composite models, where simulation strategies of different time-scales are combined, as implemented recently in E-Cell 3 and Virtual Cell [27,43]. Any mathematical model has advantages and disadvantages, and a frequent criticism of ODE modeling is its assumption that the cell is a well-stirred, deterministic reactor that disregards probabilistic fluctuation or stochasticity [2]. In more realistic cellular phenomena, where only a small number of molecules might have a role, the same starting conditions can result in different behaviors owing to stochastic effects. Therefore, a stochastic model describes each molecule as an independent, randomly colliding object. An exact stochastic method, however, comes at a huge computational cost. As stochastic methods scale to ODE methods if the number www.sciencedirect.com

of molecules is large, an intuitive solution is to apply the stochastic simulation for reactions involving small numbers of molecules, and to use ODEs for reactions involving large numbers of molecules. Thus, current simulation methods propose a dynamic switch between different granularities (e.g. scarce and abundant) [44,45]. For biotechnology applications, simulation holds great promise. A major concern of the biotechnology and pharmaceutical industry involves the improvement of yields, while decreasing fabrication costs. At this level, as in other engineering-based fields where simulation is now standard procedure, we can imagine that biological simulation will be of great use. It should facilitate the in silico testing of genetically and/or metabolically engineered organisms and greatly enhance our ability to achieve the desired objectives.

Conclusions The complex challenges of exhaustively describing and simulating cellular activities are not insurmountable, but represent a formidable task that only emerging computational and experimental methods can achieve. In addition, the largely untapped potential value of information obtained from large-scale analyses remains uncertain, owing to the difficulty in interpreting the results [30]. Concerted efforts by researchers involved in both largescale genome-wide analysis and more targeted biological approaches will continue to be important [7,30,36]. Simulation software should support such concerted efforts through static analysis (e.g. visualization and graphical representation), large-scale constraint-based approaches for steady states, and the combination of ODE and stochastic approaches for dynamic modeling. Together, all these elements shape the grand challenges for biology in the 21st century.

Update Several modeling approaches (PDE, Gillespie, and cellular automata, among others) considering intracellular spatial distribution and molecular crowding issues were recently discussed and reviewed [46]. Some of these methods can be implemented into existing multialgorithm simulation platforms to achieve more sophisticated and realistic simulations. Emili and co-workers recently reported the results of a large study of protein–protein interactions in E. coli, elucidating an interaction network for 530 complexforming bait proteins [47]. A large number of novel interactions were found, providing functional information while also highlighting the difficulty in computationally predicting interactions based solely on phylogenetic profiles or gene proximity criteria. Finally, the power of model construction, simulation and experimentation for data integration and prediction was Current Opinion in Biotechnology 2005, 16:344–349

348 Systems biology

recently demonstrated in a study of sphingolipid metabolism in yeast [48]. The method enabled the authors to predict correctly the effect of enzyme gene deletion and inositol on sphingolipid metabolism. It is one of very few studies where most predictions have actually been directly experimentally tested and confirmed.

13. Soga T, Ueno Y, Naraoka H, Ohashi Y, Tomita M, Nishioka T: Simultaneous determination of anionic intermediates for Bacillus subtilis metabolic pathways by capillary electrophoresis electrospray ionization mass spectrometry. Anal Chem 2002, 74:2233-2239.

Acknowledgements

15. Soga T, Ohashi Y, Ueno Y, Naraoka H, Tomita M, Nishioka T:  Quantitative metabolome analysis using capillary electrophoresis mass spectrometry. J Proteome Res 2003, 2:488-494. An exhaustive survey of polar metabolites using a novel and powerful CEMS platform. The combination of three different analytical configurations allowed the simultaneous detection and quantification of more than 1600 metabolites. Probably the first large-scale quantitative metabolome analysis that also shows how it can be applied for characterizing biological events such as bacterial sporulation.

The authors wish to thank Takaaki Nishioka, Tomoyoshi Soga, Tomoya Baba and Nobuyoshi Ishii (Keio University) for discussion and suggestions on the manuscript. This work was supported by (i) MEXT, Grant-in-Aid for Scientific Research on Priority Areas ‘Genome information science’, Grant-in-aid for the 21st century Center of Excellence (COE) program ‘Understanding and control of life’s function via systems biology’, and Leading project for biosimulation; (ii) METI, NEDO project ‘Development of a Technological Infrastructure for Industrial Bioprocess Project’; and (iii) the Japan Science and Technology Agency (JST). MA is a Precursory Research for Embryonic Science and Technology, Japan Science and Technology Agency, (PRESTO-JST) funded investigator.

References and recommended reading Papers of particular interest, published within the annual period of review, have been highlighted as:  of special interest  of outstanding interest

14. Soga T, Kakazu Y, Robert M, Tomita M, Nishioka T: Qualitative and quantitative analysis of amino acids by capillary electrophoresis-electrospray ionization-tandem mass spectrometry. Electrophoresis 2004, 25:1964-1972.

16. Jia L, Liu BF, Terabe S, Nishioka T: Two-dimensional separation method for analysis of Bacillus subtilis metabolites via hyphenation of micro-liquid chromatography and capillary electrophoresis. Anal Chem 2004, 76:1419-1428. 17. Aharoni A, Ric de Vos CH, Verhoeven HA, Maliepaard CA, Kruppa G, Bino R, Goodenowe DB: Nontargeted metabolome analysis by use of Fourier transform ion cyclotron mass spectrometry. OMICS 2002, 6:217-234. 18. Goodacre R, Vaidyanathan S, Bianchi G, Kell DB: Metabolic profiling using direct infusion electrospray ionisation mass spectrometry for the characterisation of olive oils. Analyst 2002, 127:1457-1462.

1.

Ishii N, Robert M, Nakayama Y, Kanai A, Tomita M: Toward largescale modeling of the microbial cell for computer simulation. J Biotechnol 2004, 113:281-294.

2.

Eungdamrong NJ, Iyengar R: Modeling cell signaling networks. Biol Cell 2004, 96:355-362.

3.

Sharom JR, Bellows DS, Tyers M: From large networks to small molecules. Curr Opin Chem Biol 2004, 8:81-90.

4.

Vidal M: A biological atlas of functional maps. Cell 2001, 104:333-339.

20. Allen J, Davey HM, Broadhurst D, Heald JK, Rowland JJ, Oliver SG, Kell DB: High-throughput classification of yeast mutants for functional genomics using metabolic footprinting. Nat Biotechnol 2003, 6:692-696.

5.

Selinger DW, Wright MA, Church GM: On the complete determination of biological systems. Trends Biotechnol 2003, 21:251-254.

21. Herrgard MJ, Covert MW, Palsson BO: Reconciling gene expression data with known genome-scale regulatory network structures. Genome Res 2003, 13:2423-2434.

6.

Tao Y, Liu Y, Friedman C, Lussier YA: Information visualization techniques in bioinformatics during the postgenomic era. Drug Discovery Today BIOSILICO 2004, 2:237-245.

22. Luscombe NM, Babu MM, Yu H, Snyder M, Teichmann SA, Gerstein M: Genomic analysis of regulatory network dynamics reveals large topological changes. Nature 2004, 431:308-312.

7.

Ideker T, Lauffenburger D: Building with a scaffold: emerging strategies for high- to low-level cellular modeling. Trends Biotechnol 2003, 21:255-262.

23. Hirai MY, Yano M, Goodenowe DB, Kanaya S, Kimura T,  Awazuhara M, Arita M, Fujiwara T, Saito K: Integration of transcriptomics and metabolomics for understanding of global responses to nutritional stresses in Arabidopsis thaliana. Proc Natl Acad Sci USA 2004, 101:10205-10210. An interesting demonstration of the potential of combined transcriptome and metabolome data integration. The authors uncovered the overall changes in the Arabidopsis metabolic network that underlie the adaptation to sulfur and nitrogen deficiencies, which would not have been possible without this two-level analysis.

8. 

Arita M: In silico atomic tracing by substrate-product relationships in Escherichia coli intermediary metabolism. Genome Res 2003, 13:2455-2466. The database provides atomic correspondences for >2000 biotransformations. Using this information, any carbon or nitrogen atom can be computationally traced in bacterial metabolism.

9. 

Pharkya P, Burgard AP, Maranas CD: OptStrain: a computational framework for redesign of microbial production systems. Genome Res 2004, 14:2367-2376. The database housing >5700 known biotransformations can hypothesize stoichiometrically balanced networks using registered reactions. The system is effective not only for metabolic reconstruction, but for bioprocess engineering. 10. Kohn KW: Molecular interaction map of the mammalian cell cycle control and DNA repair systems. Mol Biol Cell 1999, 10:2703-2734. 11. Soga T, Heiger DN: Amino acid analysis by capillary electrophoresis electrospray ionization mass spectrometry. Anal Chem 2000, 72:1236-1241. 12. Soga T, Ueno Y, Naraoka H, Matsuda K, Tomita M, Nishioka T: Pressure-assisted capillary electrophoresis electrospray ionization mass spectrometry for analysis of multivalent anions. Anal Chem 2002, 74:6224-6229.

Current Opinion in Biotechnology 2005, 16:344–349

19. Goodacre R, York EV, Heald JK, Scott IM: Chemometric discrimination of unfractionated plant extracts analyzed by electrospray mass spectrometry. Phytochemistry 2003, 62:859-863.

24. Weckwerth W, Loureiro ME, Wenzel K, Fiehn O: Differential metabolic networks unravel the effects of silent plant phenotypes. Proc Natl Acad Sci USA 2004, 101:7809-7814. 25. Huh WK, Falvo JV, Gerke LC, Carroll AS, Howson RW,  Weissman JS, O’Shea EK: Global analysis of protein localization in budding yeast. Nature 2003, 425:686-691. This study goes beyond previous similar attempts and defines the subcellular localization for a majority of yeast proteins (75%), uncovering the link between localization and transcriptional co-expression. Moreover, it provides strong additional support to the concept that proteins that colocalize have a much higher chance of interacting. 26. Ghaemmaghami S, Huh WK, Bower K, Howson RW, Belle A,  Dephoure N, O’Shea EK, Weissman JS: Global analysis of protein expression in yeast. Nature 2003, 425:737-741. A truly global survey of quantitative protein expression in yeast during logphase growth. The analysis is based on quantitative western blotting of genome-level tagged ORFs using tandem affinity purification (TAP) www.sciencedirect.com

Cell simulation Arita, Robert and Tomita 349

technology. The results also highlight the bias, in previous studies, against low abundance proteins. 27. Slepchenko BM, Schaff JC, Macara I, Loew LM: Quantitative cell  biology with the Virtual Cell. Trends Cell Biol 2003, 13:570-576. Virtual Cell is a general purpose simulation platform with user-friendly graphical interfaces. The full support of a PDE solver is impressive. 28. Gygi SP, Rochon Y, Franza BR, Aebersold R: Correlation between protein and mRNA abundance in yeast. Mol Cell Biol 1999, 19:1720-1730. 29. Tian Q, Stepaniants SB, Mao M, Weng L, Feetham MC, Doyle MJ, Yi EC, Dai H, Thorsson V, Eng J et al.: Integrated genomic and proteomic analyses of gene expression in mammalian cells. Mol Cell Proteomics 2004, 3:960-969. 30. Hughes TR, Robinson MD, Mitsakakis N, Johnston M: The  promise of functional genomics: completing the encyclopedia of a cell. Curr Opin Microbiol 2004, 7:546-554. An in-depth analysis of the power, and also the pitfalls, of large-scale functional gene analysis. This review summarizes findings from various studies in yeast and presents the comparisons using informative data transformation and visualization. 31. Vilar JM, Guet CC, Leibler S: Modeling network dynamics: the lac operon, a case study. J Cell Biol 2003, 161:471-476. 32. Henson MA: Dynamic modeling of microbial cell populations. Curr Opin Biotechnol 2003, 14:460-467. 33. Mhaskar P, Hjortso MA, Henson MA: Cell population modeling and parameter estimation for continuous cultures of Saccharomyces cerevisiae. Biotechnol Prog 2002, 18:1010-1026. 34. Korobkova E, Emonet T, Vilar JM, Shimizu TS, Cluzel P: From molecular noise to behavioural variability in a single bacterium. Nature 2004, 428:574-578. 35. Castrillo JI, Oliver SG: Yeast as a touchstone in post-genomic research: strategies for integrative analysis in functional genomics. J Biochem Mol Biol 2004, 37:93-106. 36. Mori H: From the sequence to cell modeling: comprehensive functional genomics in Escherichia coli. J Biochem Mol Biol 2004, 37:83-92. 37. Kang Y, Durfee T, Glasner JD, Qiu Y, Frisch D, Winterberg KM, Blattner FR: Systematic mutagenesis of the Escherichia coli genome. J Bacteriol 2004, 186:4921-4930.

www.sciencedirect.com

38. Proudfoot M, Kuznetsova E, Brown G, Rao NN, Kitagawa M, Mori H, Savchenko A, Yakunin AF: General enzymatic screens identify three new nucleotidases in Escherichia coli: biochemical characterization of SurE, YfbR, and YjjG. J Biol Chem 2004, 279:54687-54694. 39. Ge H, Walhout AJ, Vidal M: Integrating ‘omic’ information: a bridge between genomics and systems biology. Trends Genet 2003, 19:551-560. 40. Ibarra RU, Edwards JS, Palsson BO: Escherichia coli K-12 undergoes adaptive evolution to achieve in silico predicted optimal growth. Nature 2002, 420:186-189. 41. Fong SS, Palsson BO: Metabolic gene-deletion strains of Escherichia coli evolve to computationally predicted growth phenotypes. Nat Genet 2004, 36:1056-1058. 42. Covert MW, Knight EM, Reed JL, Herrgard MJ, Palsson BO:  Integrating high-throughput and computational data elucidates bacterial networks. Nature 2004, 429:92-96. In this truly large-scale approach, regulatory and metabolic networks of E. coli are computationally reconstructed and reconciled with experimentally derived (>13 000) growth phenotypes. The in silico model could predict the outcomes of almost 80% of the gene-knockout experiments. 43. Takahashi K, Kaizu K, Hu B, Tomita M: A multi-algorithm, multitimescale method for cell simulation. Bioinformatics 2004, 20:538-546. 44. Puchalka J, Kierzek AM: Bridging the gap between stochastic and deterministic regimes in the kinetic simulations of the biochemical reaction networks. Biophys J 2004, 86:1357-1372. 45. Vasudeva K, Bhalla US: Adaptive stochastic-deterministic chemical kinetic simulations. Bioinformatics 2004, 20:78-84. 46. Takahashi K, Arjunan SN, Tomita M: Space in systems biology of signaling pathways — towards intracellular molecular crowding in silico. FEBS Lett 2005, 579:1783-1788. 47. Butland G, Peregrin-Alvarez JM, Li J, Yang W, Yang X, Canadien V, Starostine A, Richards D, Beattie B, Krogan N et al.: Interaction network containing conserved and essential protein complexes in Escherichia coli. Nature 2005, 433:531-537. 48. Alvarez-Vasquez F, Sims KJ, Cowart LA, Okamoto Y, Voit EO, Hannun YA: Simulation and validation of modelled sphingolipid metabolism in Saccharomyces cerevisiae. Nature 2005, 433:425-430.

Current Opinion in Biotechnology 2005, 16:344–349