MetaReg: a platform for modeling, analysis and visualization of biological systems using largescale experimental data
 Igor Ulitsky†^{1},
 Irit GatViks†^{1, 2} and
 Ron Shamir^{1}Email author
DOI: 10.1186/gb200891r1
© Ulitsky et al.; licensee BioMed Central Ltd. 2008
Received: 4 July 2007
Accepted: 2 January 2008
Published: 02 January 2008
Abstract
MetaReg http://acgt.cs.tau.ac.il/metareg/application.html is a computational tool that models cellular networks and integrates experimental results with such models. MetaReg represents established knowledge about a biological system, available today mostly in informal form in the literature, as probabilistic network models with underlying combinatorial regulatory logic. MetaReg enables contrasting predictions with measurements, model improvements and studying whatif scenarios. By summarizing prior knowledge and providing visual and computational aids, it helps the expert explore and understand her system better.
Rationale
Given the recent accumulation of high throughput biological data, the task of integrating and analyzing largescale datasets is a major challenge. A variety of computational modeling approaches have been developed for the analysis of such datasets, such as clustering [1, 2] and topological interaction network models [3, 4]. While these approaches give a broad, low resolution picture of cellular processes, many biologists are interested in a specific subsystem, and wish to use the results from experiments in order to refine the current knowledge on the system. This analysis of data in the context of the available knowledge is often performed in an informal manner: The researcher sketches a diagram of a relevant subsystem according to the current knowledge. This diagram summarizes and organizes the available knowledge, and assists the expert in analyzing the predicted state of the system in various possible experiments. The predictions are then compared to experimental measurements, and if a discrepancy is found, additional experiments are performed, and the diagram is iteratively refined.
In the case of complex biological systems and massive amounts of data, manual construction of the model, state predictions, comparison with data and systematic model refinements are impractical, and automatic computational methodologies must be employed [5, 6]. To address the need for such an analysis workflow, we developed MetaReg, an integrative tool for analysis of steadystate, highthroughput data in the context of specific biological systems. The theoretical foundations of the MetaReg methodology and algorithms are outlined below in the 'MetaReg's algorithmic layer' section (for a complete description, see [7]). While making some gross simplifying assumptions about the behavior of real biological systems, the model was demonstrated to be highly effective on several systems [7–9]. MetaReg enables easy conversion of the current qualitative knowledge on a particular subsystem into a mathematical model, including logical relations among the biological components. The system is represented by a probabilistic graphical model called a Bayesian network [10], which allows distinguishing between regulatory relations that are known at a high level of certainty and those that are more speculative. Given the model, MetaReg predicts the level of each variable under any given genetic perturbation or environmental stimuli. Moreover, MetaReg allows incorporation of high throughput data, and graphical comparison between model predictions and measurements. The most advanced MetaReg capability is suggesting model refinements by systematically seeking changes that increase the fit between model predictions and experimental measurements.
The MetaReg application
MetaReg core functionality
Model construction
The first step in utilizing MetaReg is the construction of the biological system model on the 'model canvas' (Figure 2a). A MetaReg model consists of a set of biological variables and their regulatory logics. The 'variables' represent different biological entities (mRNA, protein, metabolite, and so on). Each variable may attain several discrete 'states' (three states by default), representing, for example, the transcript level of an mRNA, or the activity level of an enzyme. The state of a variable v is influenced by the states of the variables that are connected to v by incoming edges. These variables are called the 'regulators' of v. Most importantly, every variable is assigned a discrete 'logic', which defines its state given the states of its regulators. For example, if variable A has two activators B and C, its logic might be Max(B, C). We assume all the logics represent steadystate regulatory relations, and thus the model represents the steadystate behavior of the biological system. Every logic is associated with a probability that indicates the certainty in the prior biological knowledge. For example, if a logic is known with high certainty, it will be assigned with a high probability (for example, 90%), and alternative logics will have low probabilities.
The application offers several tools to help in model construction. Variables can be selected from and automatically linked to known databases, such as SGD [12] and NCBI Gene [13] (Figure 2b). Each variable can be attributed with links to relevant journal publications from PubMed, enabling further model curation. The application provides several gadgets for logic definition, including scripting, a tabular editor and a logic wizard (Figure 2c) for hierarchical construction of complex logics. The type of each regulation, activation (→), repression (⊣) or other (○) is automatically deduced based on the logic of the regulatee (the regulated variable). The model canvas is fully interactive, including capabilities for manual or automatic variable positioning and highlighting of different sets of variables, such as all the metabolites or all the cycles in the model.
Model simulation
In order to view the behavior of the model in response to different experimental treatments, simulations can be performed. Given a particular experimental treatment, the possible system states are computed as described in [8]. A 'system state' is an assignment of states to all the variables in the model. The user can dynamically design an experimental treatment scenario and visually analyze the system state on the model canvas. If the model contains cycles, several system states might be feasible, and the user can navigate among them.
Data integration
The application can integrate 'observations' (measurements) from multiple studies. The measured biological components are automatically matched to the model variables. For example, gene expression data are automatically matched to the corresponding mRNA variables, and protein measurements are matched to the corresponding protein variables. As part of the data import, the user must specify the 'experimental treatment' used in each experiment, including the environmental stimulations and genetic perturbations performed in each particular experiment. For example, if the experiment was performed in surplus of nitrogen and on a yeast strain where Leu3 is knocked out, the experimental treatment is 'Leu3 = 0; Nitrogen = 2', where Leu3 and Nitrogen are model variables. Once the data are imported, it is possible to visualize all measured variables under each of the experiments in a single data matrix (Figure 2d; see below), or to view the measurements of a specific experiment projected on the model canvas (Figure 2a).
Comparing predictions with observations
In order to evaluate the model, the 'predicted' levels of each variable are compared to its 'observed' levels under each experiment. MetaReg provides a prediction engine that infers probabilistically the expected level of each variable in each experiment, given the network model and the experimental treatment (see [7]). MetaReg supports two visualization tools to compare these predictions with the observations, both designed to highlight cases of discrepancies, which are often the starting point of further research. First, the observed and the predicted values for a single experiment can be projected side by side on the model canvas (Figure 2a). The second visualization tool provides a comprehensive view of the discrepancies across all the experiments, in which each cell contains colorcoded representation of the observed and the predicted values, along with a representation of the discrepancy between them (Figure 2d). This view allows simple detection of discrepancy 'hotspots' in which the model fails to explain the data.
Model refinement
Our methodology enables refinement of the model to obtain better fit between model predictions and observations. The input of the refinement process is the target variable and a set of regulators. MetaReg searches among all possible regulatory logics and outputs the most significant one. The suggested logic can be further edited by the user (Figure 2e). This way the user can test hypotheses about variable regulation.
Case study: leucine biosynthesis in Saccharomyces cerevisiae
Modeling and simulations
We present a model for leucine biosynthesis and related signaling pathways in Saccharomyces cerevisiae. Building on literature reports, we constructed a detailed model of known regulatory relations in this system. The model contains 47 variables (nodes) and 67 regulations (arcs). The model is available from our web site [14].
Leucine is an essential branchedchain amino acid generated from pyruvate via αketoisovalerate, αisopropylmalate (αIPM) and βIPM in a linear pathway in which nine catalyzing enzymes are involved (Ilv2, Ilv3, Ilv5, Leu9, Leu4, Leu1, Leu2, Bat1, Bat2). The regulation of leucine production is controlled by several known mechanisms [15].
Several leucine biosynthetic enzymes are subject to transcription regulation via the general regulatory pathway of amino acid biosynthesis. Starvation for any amino acid induces the translation of Gcn4 via Gcn2. Gcn4 is a transcriptional activator of enzymes that catalyze several amino acid biosynthesis pathways, including the leucine biosynthetic pathway.
The control of several catalyzing enzymes is regulated by the transcriptional activator Leu3. The activity of Leu3 is regulated by αIPM, an intermediate of the pathway acting as a coinducer. When αIPM is present, Leu3 acts as activator; when αIPM is absent, Leu3 acts as repressor [15]. Hence, αIPM serves as a sensor of leucine production.
The enzymatic activity of Leu4 is subject to two major controls by metabolites. The first is feedback (end product) inhibition by leucine. At high levels of leucine, Leu4 activity is inhibited, and causes a reduction in the production of the pathway. The second control is inactivation by coenzyme A, a product of the reaction catalyzed by Leu4 and a central energy metabolite in the mitochondria. This control serves as a link between the metabolic process and the energy metabolism context.
We used three states for each mRNA variable: state '0' represents reduced transcription level compared to the wild type, state '1' represents the wildtype transcription level when cells are grown on YPD medium, and state '2' represents increased transcription level. Similarly, each protein has three states reflecting its activity level (high = '2', medium = '1', low = '0'). The modeling of Leu3 is a special case, since we had to represent its dual role as activator and repressor. We used state '0' for its repressive mode, state '1' represents no effect (for example, in the leu3 mutant), and state '2' indicates the Leu3 activator mode. For example, a simulation of the system behavior in leucine starvation is shown in Figure 3.
Data preparation
We integrated expression profiles from four datasets that contain treatments pertinent to our model: seven profiles in rapamycin treatment after 15, 30, 60 and 120 minutes of incubation and in amino acid deprivation after 1, 1.5 and 2 hours of incubation [16]; six profiles in histidine starvation and various Gcn4 perturbations [17]; six profiles of chemostat growth in nitrogen limiting conditions with and without Leu3 perturbation [18]; and six profiles in nitrogen depletion after 8, 12 and 24 hours of treatment and in amino acid and adenine starvation after 1, 2 and 4 hours of treatment [19]. A complete description of the profiles, the experimental treatments under which they were obtained and the data preprocessing, is available in Additional data file 1.
Evaluation of the model in accordance with data
The leucine biosynthetic genes LEU1, LEU2, LEU4 and BAP2 show an unexpected decrease in expression in the leu3 mutant strain (Figure 4b, columns 1618). The reduction was surprising since Leu3 is known to act as a repressor in these experiments.
In gcn4 mutant strains, we observed an increase in the mRNA levels of the leucine biosynthetic genes BAT1, ILV2, ILV3 and ILV5 following 3AT treatment (histidine starvation; Figure 4a, columns 1112). In our model the effect of general amino acid control on these genes is mediated solely by Gcn4. Since Gcn4 is absent in these experiments, our model does not predict such an increase, and a discrepancy appears (Figure 4b, column 12).
For LEU3, we observed an increase in expression in two gcn4 mutant strains and in nitrogen limitation experiments (Figure 4b, row 11, columns 1115). According to the literature, LEU3 mRNA is upregulated by either Gcn4 or Leu3 TFs. As no amino acid shortage occurs in these experiments, neither Gcn4 nor Leu3 are expected to be active, hence the model predicts a low level of LEU3 mRNA, in contradiction to the observed increase.
Following a rapamycin treatment, we observed a consistent decrease in the levels of four biosynthetic genes, BAT1, ILV3, ILV5 and LEU1. The effect of rapamycin on the biosynthetic genes is known to be mediated by the TOR pathway through Gcn4 [20]. It is thus expected that under rapamycin treatment, Gcn4 will be active, while Leu3 will not be active. Consequently, the levels of the leucine biosynthesis genes (LEU1, ILV3, ILV5, BAT1) regulated by Gcn4 should be alleviated. Surprisingly, we witness a downregulation of these genes.
For LEU9, BAT2, BAP3 and TAT1, we could not find any report on their regulation in the literature, and thus their predicted level is constant. Hence, the discrepancies merely reflect the lack of knowledge about them.
Leucine model refinement
In order to improve the fit of the model's predictions to the observed data, we used MetaReg's refinement algorithm. We focus here on two representative examples of model refinement. In these examples we suggest improved logics for the way in which Leu3 and Gcn4 jointly regulate LEU9, BAT2 and LEU2.
LEU2 expression is known to be affected only by Leu3 [15]. Indeed, the suggested logic (Figure 5) shows that the state of Gcn4 does not influence Leu2. As expected, when Leu3 should act as activator (Leu3 = '2') there is high transcription (LEU2 = '2'). However, we do not detect the expected repressive effect of Leu3 on its targets. When Leu3 should act as repressor (Leu3 = '0'), we observe medium LEU2 transcription (LEU2 = '1') instead of the expected low transcription.
Figure 5b,c illustrates the refinement process. During refinement, MetaReg tests the predicted activity levels of the TFs (Gcn4 and Leu3) against the observed level of the mRNA in each experiment (Figure 5b), and computes the best logic between the regulators' predicted level and the observations. Consequently, the discrepancies observed for LEU2 and LEU9 in our initial model (before refinement) are drastically reduced after refinement (Figure 5c).
In the case of LEU1, BAT1, LEU4, ILV2, ILV3 and ILV5, the results were similar to LEU2 (not shown). For BAP2, BAP3, TAT1 and LEU3, MetaReg did not succeed in deriving a high confidence logical relation, due to inconsistent effects that could not be explained by the model. For example, for TAT1, only downregulation is observed in the data (Figure 4a, last row). For BAP3, we observe an inconsistency between two sets of nitrogen depletion experiments in different studies (Figure 4a, columns 1315 versus 1921). This probably indicates that each of those genes is regulated by additional elements that are not included in the model.
MetaReg's algorithmic layer
In this section, we briefly outline the algorithmic layer behind the MetaReg application. A full description can be found in [7].
Modeling prior knowledge
Our model consists of variables X_{1}...X_{n}, represented by nodes, and regulations among them, represented by arcs. The set of variables that together regulate variable X_{i} are called its 'regulatory unit', denoted Pa_{i}. This is the set of nodes that have arcs directed into X_{i}. Each variable can be in one of several discrete 'states', and its state in any condition is assumed to be determined by its 'logic', that is, a discrete function of its regulators' states in that condition. Note that this assumption implies that the relevant conditions are in steady state. In order to model our confidence in the prior knowledge, the logic of a variable X_{i} is formulated probabilistically as our level of certainty that the variable attains a certain state given the state of its regulatory unit. The uncertainty is modeled by the conditional probability θ^{i}(X_{i}  Pa_{i}). This approach allows us to distinguish between regulatory logics that are known at a high level of certainty and those that are more speculative.
The experimental treatment is modeled by fixing the states of each variable that correspond to the environment, and by changing the regulation function priors to reflect the perturbations (for example, when a gene is knocked out, its level is set to zero under that condition, irrespective of the levels of its regulators).
Data integration
In practice, biological measurements are continuous, and one does not know in advance how to translate them into discrete states. To overcome this, each logical variable X_{i} is associated with an observed realvalued variable Y_{i}, and the conditional distribution ψ^{i}(Y_{i}  X_{i}) specifies the probability of the variable Y_{i} to attain a certain observed real value given its state. Hence, ψ^{i}(Y_{i}  X_{i}) translates the actual measurements into the discrete model without applying any a priori discretization to the data. In MetaReg, each ψ is modeled as a mixture of Gaussians.
The complete computational model
where Z is a normalization constant. The conditional probabilities θ^{i} are known from our prior knowledge of the biological system, and ψ are determined by maximizing a likelihood score using an ExpectationMaximization procedure. This model corresponds to a Bayesian network in the case of acyclic dependencies, or to a factor graph in the more general case where the model contains feedback loops.
Computing model predictions
The 'predicted level' is the expected value of a variable X_{i} given the model and the experimental procedure applied. This is obtained by first computing the posterior states distribution of X_{i} using a standard probabilistic inference method called Loopy Belief Propagation [21]. This way we obtain a probabilistic average of all its possible system modes. Then, the (continuous) predicted level of X_{i} is its expectation given θ^{i} and its states distribution. The comparison of predicted and observed levels (both on the model canvas and in a discrepancy matrix) displays both levels as real values.
Logic refinement
Given a target gene and its candidate regulatory unit, the refinement process searches in the space of discrete regulatory logics in order to achieve a logic with a locally maximum Bayesian score, while fixing the logics of all other variables. Due to an exponential number of possible logics, we apply a greedy heuristic. In the case of ties the algorithm chooses randomly among the equally scored improvements. The ψ^{i} parameters depend strongly on the particular model logics, and thus we reoptimize them using an expectationmaximization (EM)like procedure during each step of the logical refinement procedure. Note that the refinement process utilizes the Loopy Belief Propagation algorithm, and thus the solution builds on probabilistic averaging of all possible system modes.
Discussion
Available tools related to MetaReg
Tool type  Tools  Description  Relation to MetaReg 

Network or model visualization tools  Cytoscape [28] Visant [29] CellDesigner [30] Reviewed in [31]  Tools for constructing visualizations of interaction and regulatory networks. These networks can then be integrated with highthroughput data  These tools offer powerful visualization aids and other analysis aids, but they do not address regulatory logics and do not offer model evaluation or refinement mechanisms 
Kinetic and continuous modeling tools  Gepasi [32] BioNetS [33] Dynetica [34] PyBioS [35] Reviewed in [3638]  Tools allowing detailed dynamical modeling with kinetic parameters and differential equations  These tools can perform detailed model analysis by accurate dynamical simulations, but they cannot discover new mechanisms and rely on detailed mechanistic understanding of the system^{1} 
Logical modeling tools  BIOCHAM [39] Bionet [40] CellNetAnalyzer [41] GINsim [42]  Tools for modeling regulatory systems using various formalisms, for example, Boolean, discrete, fuzzy logic and so on  Allow model evaluation through simulations, but are not designed for model evaluation and refinement in accordance with high throughput data 
MetaReg is implemented efficiently, computing predictions and logic refinements within a few seconds for 100 nodes, and within an hour for 6,000 nodes (using a network with no more than three regulators per variable, 90% certainty level in all logics, and 100 gene expression profiles). However, the model has practical size limitations: the prediction algorithm runtime increases exponentially with the average number of regulators per variable. Also, for large models with over 300 variables, the automatic layout of the model topology may take several minutes.
MetaReg formalizes the biological system using discrete component states, assuming that the system is in steady state. Clearly these crucial assumptions are a simplification of the biological reality. By making such assumptions, we tried to strike a practical balance between our wish to enable a faithful description of the biological system and the scarcity of accurate knowledge at very high resolution. Indeed, biological processes are inherently temporal, but when the sampling rate (the number and time resolution of experiments) is low relative to the rate of the regulatory mechanisms, we believe that our results here as well as in [7–9] show that the steady state assumption is reasonable.
The accuracy of the prediction and refinement processes may be sensitive to the model size and the certainty in the logics. We have shown previously that the algorithms are highly robust to certainty level on small networks [7]. Indeed, the results shown in the leucine example were obtained using a uniform certainty level of 0.99 for all variables, but we obtained very similar results when using certainty levels of 0.95 and 0.9 (not shown). However, the robustness of our methods to model size and to certainty levels requires further systematic exploration.
A major prerequisite to using MetaReg is formalizing high quality prior knowledge on the pathway of interest. Several efforts to generate databases of curated knowledge on signaling pathway are currently under way (for example, BioModels [22], Reactome [23] and SPIKE [24]). Thanks to such efforts, it will soon be relatively easy to apply the MetaReg methodology in studying many additional biological systems.
Availability and requirements
Project name: MetaReg (home page at [25]).
Operating system(s): Windows.
Programming language: Java for the envelope and C++ for the algorithms.
Other requirements: Java 1.5 or higher.
License: free for noncommercial users.
Any restrictions to use by nonacademics: License needed.
Additional data files
The following additional data are available with the online version of this paper. Additional data file 1 provides a complete description of the profiles, the experimental treatments under which they were obtained and the data preprocessing.
Notes
Abbreviations
 EM:

expectation maximization
 IPM:

isopropylmalate
 TF:

transcription factor.
Declarations
Acknowledgements
This work was supported by the EMICD project, which is funded by the European Commission within its FP6 Programme, under the thematic area 'Life sciences, genomics and biotechnology for health', contract number LSHGCT2003503269. The information in this document is provided as is and no guarantee or warranty is given that the information is fit for any particular purpose. The user thereof uses the information at its sole risk and liability. The graphical capabilities of the model layout and related dialogs are deeply based on the implementations of the PIVOT [26] and SPIKE [24] software developed in close collaboration by R Shamir's group and Y Shiloh's group at the School of Medicine, Tel Aviv University. We are especially indebted to Giora Sternberg and Ran Blekhman for their fruitful support. The interactive data matrix display has been developed by Israel Steinfeld for the SIMBA website [27]. We also would like to thank Amos Tanay for helpful discussions and Ewa Szczurek for testing MetaReg and helping to improve it. I Ulitsky is a fellow of the Edmond J Safra Bioinformatics Program at TelAviv University.
Authors’ Affiliations
References
 Eisen MB, Spellman PT, Brown PO, Botstein D: Cluster analysis and display of genomewide expression patterns. Proc Natl Acad Sci USA. 1998, 95: 1486314868. 10.1073/pnas.95.25.14863.PubMedPubMed CentralView ArticleGoogle Scholar
 Segal E, Shapira M, Regev A, Pe'er D, Botstein D, Koller D, Friedman N: Module networks: identifying regulatory modules and their conditionspecific regulators from gene expression data. Nat Genet. 2003, 34: 166176.PubMedView ArticleGoogle Scholar
 Bader GD, Hogue CW: An automated method for finding molecular complexes in large protein interaction networks. BMC Bioinformatics. 2003, 4: 210.1186/1471210542.PubMedPubMed CentralView ArticleGoogle Scholar
 Rives AW, Galitski T: Modular organization of cellular networks. Proc Natl Acad Sci USA. 2003, 100: 11281133. 10.1073/pnas.0237338100.PubMedPubMed CentralView ArticleGoogle Scholar
 Klipp E, Nordlander B, Kruger R, Gennemark P, Hohmann S: Integrative model of the response of yeast to osmotic shock. Nat Biotechnol. 2005, 23: 975982. 10.1038/nbt1114.PubMedView ArticleGoogle Scholar
 Covert MW, Knight EM, Reed JL, Herrgard MJ, Palsson BO: Integrating highthroughput and computational data elucidates bacterial networks. Nature. 2004, 429: 9296. 10.1038/nature02456.PubMedView ArticleGoogle Scholar
 GatViks I, Tanay A, Raijman D, Shamir R: A probabilistic methodology for integrating knowledge and experiments on biological networks. J Comput Biol. 2006, 13: 165181. 10.1089/cmb.2006.13.165.PubMedView ArticleGoogle Scholar
 GatViks I, Tanay A, Shamir R: Modeling and analysis of heterogeneous regulation in biological networks. J Comput Biol. 2004, 11: 10341049. 10.1089/cmb.2004.11.1034.PubMedView ArticleGoogle Scholar
 GatViks I, Shamir R: Refinement and expansion of signaling pathways: the osmotic response network in yeast. Genome Res. 2007, 17: 358367. 10.1101/gr.5750507.PubMedPubMed CentralView ArticleGoogle Scholar
 Pearl J: Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. 1988, Morgan Kaufmann, Palo Alto, CAGoogle Scholar
 MetaReg Manual. [http://acgt.cs.tau.ac.il/metareg/manual/]
 Cherry JM, Adler C, Ball C, Chervitz SA, Dwight SS, Hester ET, Jia Y, Juvik G, Roe T, Schroeder M, et al: SGD: Saccharomyces Genome Database. Nucleic Acids Res. 1998, 26: 7379. 10.1093/nar/26.1.73.PubMedPubMed CentralView ArticleGoogle Scholar
 Maglott D, Ostell J, Pruitt KD, Tatusova T: Entrez Gene: genecentered information at NCBI. Nucleic Acids Res. 2005, D5458. 33 Database
 MetaReg Application. [http://acgt.cs.tau.ac.il/metareg/application.html]
 Kohlhaw GB: Leucine biosynthesis in fungi: entering metabolism through the back door. Microbiol Mol Biol Rev. 2003, 67: 115. 10.1128/MMBR.67.1.115.2003.PubMedPubMed CentralView ArticleGoogle Scholar
 Hardwick JS, Kuruvilla FG, Tong JK, Shamji AF, Schreiber SL: Rapamycinmodulated transcription defines the subset of nutrientsensitive signaling pathways directly controlled by the Tor proteins. Proc Natl Acad Sci USA. 1999, 96: 1486614870. 10.1073/pnas.96.26.14866.PubMedPubMed CentralView ArticleGoogle Scholar
 Natarajan K, Meyer MR, Jackson BM, Slade D, Roberts C, Hinnebusch AG, Marton MJ: Transcriptional profiling shows that Gcn4p is a master regulator of gene expression during amino acid starvation in yeast. Mol Cell Biol. 2001, 21: 43474368. 10.1128/MCB.21.13.43474368.2001.PubMedPubMed CentralView ArticleGoogle Scholar
 Boer VM, Daran JM, Almering MJ, de Winde JH, Pronk JT: Contribution of the Saccharomyces cerevisiae transcriptional regulator Leu3p to physiology and gene expression in nitrogen and carbonlimited chemostat cultures. FEMS Yeast Res. 2005, 5: 885897. 10.1016/j.femsyr.2005.04.003.PubMedView ArticleGoogle Scholar
 Gasch AP, Spellman PT, Kao CM, CarmelHarel O, Eisen MB, Storz G, Botstein D, Brown PO: Genomic expression programs in the response of yeast cells to environmental changes. Mol Biol Cell. 2000, 11: 42414257.PubMedPubMed CentralView ArticleGoogle Scholar
 Rohde JR, Campbell S, ZuritaMartinez SA, Cutler NS, Ashe M, Cardenas ME: TOR controls transcriptional and translational programs via SapSit4 protein phosphatase signaling effectors. Mol Cell Biol. 2004, 24: 83328341. 10.1128/MCB.24.19.83328341.2004.PubMedPubMed CentralView ArticleGoogle Scholar
 Kschischang FR, Frey BJ, Loeliger HA: Factor graphs and the sumproduct algorithm. IEEE Trans Information Theory. 2001, 47: 498519. 10.1109/18.910572.View ArticleGoogle Scholar
 Le Novère N, Bornstein B, Broicher A, Courtot M, Donizelli M, Dharuri H, Li L, Sauro H, Schilstra M, Shapiro B: BioModels Database: a free, centralized database of curated, published, quantitative kinetic models of biochemical and cellular systems. Nucleic Acids Res. 2006, D689691. 10.1093/nar/gkj092. 34 Database
 Vastrik I, D'Eustachio P, Schmidt E, Stein L: Reactome: a knowledge base of biologic pathways and processes. Genome Biol. 2007, 8: R3910.1186/gb200783r39.PubMedPubMed CentralView ArticleGoogle Scholar
 Fraser HB, Hirsh AE, Steinmetz LM, Scharfe C, Feldman MW: Evolutionary rate in the protein interaction network. Science. 2002, 296: 750752. 10.1126/science.1068696.PubMedView ArticleGoogle Scholar
 MetaReg. [http://acgt.cs.tau.ac.il/metareg]
 Orlev N, Shamir R, Shiloh Y: PIVOT: protein interacions visualizatiOn tool. Bioinformatics. 2004, 20: 424425. 10.1093/bioinformatics/btg426.PubMedView ArticleGoogle Scholar
 Tanay A, Steinfeld I, Kupiec M, Shamir R: Integrative analysis of genomewide experiments in the context of a large highthroughput data compendium. Mol Syst Biol. 2005, 1: 2005.000210.1038/msb4100005.PubMedPubMed CentralView ArticleGoogle Scholar
 Shannon P, Markiel A, Ozier O, Baliga NS, Wang JT, Ramage D, Amin N, Schwikowski B, Ideker T: Cytoscape: a software environment for integrated models of biomolecular interaction networks. Genome Res. 2003, 13: 24982504. 10.1101/gr.1239303.PubMedPubMed CentralView ArticleGoogle Scholar
 Hu Z, Mellor J, Wu J, Yamada T, Holloway D, Delisi C: VisANT: dataintegrating visual framework for biological networks and modules. Nucleic Acids Res. 2005, W352357. 10.1093/nar/gki431. 33 Web Server
 Kitano H, Funahashi A, Matsuoka Y, Oda K: Using process diagrams for the graphical representation of biological networks. Nat Biotechnol. 2005, 23: 961966. 10.1038/nbt1111.PubMedView ArticleGoogle Scholar
 Bell GW, Lewitter F: Visualizing networks. Methods Enzymol. 2006, 411: 408421. 10.1016/S00766879(06)110228.PubMedView ArticleGoogle Scholar
 Baigent S: Software review. Gepasi 3.0. Brief Bioinform. 2001, 2: 300302. 10.1093/bib/2.3.300.PubMedView ArticleGoogle Scholar
 Adalsteinsson D, McMillen D, Elston TC: Biochemical Network Stochastic Simulator (BioNetS): software for stochastic modeling of biochemical networks. BMC Bioinformatics. 2004, 5: 2410.1186/14712105524.PubMedPubMed CentralView ArticleGoogle Scholar
 You L, Hoonlor A, Yin J: Modeling biological systems using Dynetica  a simulator of dynamic networks. Bioinformatics. 2003, 19: 435436. 10.1093/bioinformatics/btg009.PubMedView ArticleGoogle Scholar
 RodriguezNavarro S, Fischer T, Luo MJ, Antunez O, Brettschneider S, Lechner J, PerezOrtin JE, Reed R, Hurt E: Sus1, a functional component of the SAGA histone acetylase complex and the nuclear poreassociated mRNA export machinery. Cell. 2004, 116: 7586. 10.1016/S00928674(03)010250.PubMedView ArticleGoogle Scholar
 Alves R, Antunes F, Salvador A: Tools for kinetic modeling of biochemical networks. Nat Biotechnol. 2006, 24: 667672. 10.1038/nbt0606667.PubMedView ArticleGoogle Scholar
 Aldridge BB, Burke JM, Lauffenburger DA, Sorger PK: Physicochemical modelling of cell signalling pathways. Nat Cell Biol. 2006, 8: 11951203. 10.1038/ncb1497.PubMedView ArticleGoogle Scholar
 Price ND, Reed JL, Palsson BO: Genomescale models of microbial cells: evaluating the consequences of constraints. Nat Rev Microbiol. 2004, 2: 886897. 10.1038/nrmicro1023.PubMedView ArticleGoogle Scholar
 Calzone L, Fages F, Soliman S: BIOCHAM: an environment for modeling biological systems and formalizing experimental knowledge. Bioinformatics. 2006, 22: 18051807. 10.1093/bioinformatics/btl172.PubMedView ArticleGoogle Scholar
 Bosl WJ: Systems biology by the rules: hybrid intelligent systems for pathway modeling and discovery. BMC Syst Biol. 2007, 1: 1310.1186/17520509113.PubMedPubMed CentralView ArticleGoogle Scholar
 Klamt S, SaezRodriguez J, Gilles ED: Structural and functional analysis of cellular networks with CellNetAnalyzer. BMC Syst Biol. 2007, 1: 210.1186/1752050912.PubMedPubMed CentralView ArticleGoogle Scholar
 Gonzalez AG, Naldi A, Sanchez L, Thieffry D, Chaouiya C: GINsim: a software suite for the qualitative modelling, simulation and analysis of regulatory networks. Biosystems. 2006, 84: 91100. 10.1016/j.biosystems.2005.10.003.PubMedView ArticleGoogle Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.