Simultaneous model discrimination and parameter estimation in dynamic models of cellular systems
 Maria RodriguezFernandez^{1, 4},
 Markus Rehberg^{2},
 Andreas Kremling^{3} and
 Julio R Banga^{4}Email author
DOI: 10.1186/17520509776
© RodriguezFernandez et al.; licensee BioMed Central Ltd. 2013
Received: 8 May 2012
Accepted: 8 August 2013
Published: 12 August 2013
Abstract
Background
Model development is a key task in systems biology, which typically starts from an initial model candidate and, involving an iterative cycle of hypothesesdriven model modifications, leads to new experimentation and subsequent model identification steps. The final product of this cycle is a satisfactory refined model of the biological phenomena under study. During such iterative model development, researchers frequently propose a set of model candidates from which the best alternative must be selected. Here we consider this problem of model selection and formulate it as a simultaneous model selection and parameter identification problem. More precisely, we consider a general mixedinteger nonlinear programming (MINLP) formulation for model selection and identification, with emphasis on dynamic models consisting of sets of either ODEs (ordinary differential equations) or DAEs (differential algebraic equations).
Results
We solved the MINLP formulation for model selection and identification using an algorithm based on Scatter Search (SS). We illustrate the capabilities and efficiency of the proposed strategy with a case study considering the KdpD/KdpE system regulating potassium homeostasis in Escherichia coli. The proposed approach resulted in a final model that presents a better fit to the in silico generated experimental data.
Conclusions
The presented MINLPbased optimization approach for nestedmodel selection and identification is a powerful methodology for model development in systems biology. This strategy can be used to perform model selection and parameter estimation in one single step, thus greatly reducing the number of experiments and computations of traditional modeling approaches.
Keywords
Dynamic modelling Parameter estimation Model discrimination Global optimizationBackground
Model development is a key task in systems biology, and involves different steps, such as model calibration, experimental design and model refinement which usually take place in an iterative way (see reviews in [1–5]). The process of building a model of a biological system typically starts by generating an initial model candidate, or by taking one from the preexisting knowledge, and then involves an iterative cycle of hypothesesdriven model modifications, new experimentation and subsequent model identification steps, finally leading to a satisfactory refined model [6, 7]. Thus, model selection, experimentation and model refinement can be considered the basic elements of systems biology [8].
A number of researches have proposed different iterative schemes for model development involving the steps of parameter estimation, identifiability analysis, and optimal experimental design [9–12]. The related topic of optimal experimental design for parameter estimation [3, 13] and for model discrimination [14–16] is receiving increased attention in recent years. Lillacci and Khammash [17] introduced a new method for parameter estimation based on Kalman filtering that can also be used to discriminate among alternate models of the same biological process.
Verheijen [18] presented an overview of model selection practices, highlighting the main criteria for choosing out of a large set of models: level of rigor, accuracy with respect to data, adequacy of the model, and its flexibility and computational complexity. He also identifies developments in optimizationbased approaches [19, 20] as very promising, but recognizing its limitations due to numerical and algorithmic challenges. Although research along this line has continued [21, 22], it still remains as a challenging numerical problem.
Here, we present a method to simultaneously select a model and calibrate it in a single step. This contribution is based on the following four key ideas: (i) frequently, iterative model development cycles can be considered in a more compact way if sets of hypotheses can be grouped together and formulated as a parameterized set of models, from which the best alternative must be selected; (ii) we consider the problem of model selection formulating it as a simultaneous model selection and parameter identification problem; (iii) further, in order to make the selection decision in a systematic way, we formulate it as an optimization problem [23] acting on the parameterized set of models; (iv) the optimization problem, which belongs to the class of mixedinteger nonlinear programming (MINLP) problems, is solved by recently developed algorithms based on metaheuristics.
The paper is organized as follows: First, we describe the framework used for model selection and identification, based on the nested models paradigm. Then we state the corresponding optimization problem using a formulation based on mixedinteger nonlinear programming subject to differential and algebraic constraints. In the following sections, we describe the application of this methodology to a case study considering a dynamic model of the KdpD/KdpE system of Escherichia coli. Finally, we provide a discussion of the results and summarize the main conclusions of the study.
Methods
To the best of our knowledge, this is the first time that an MINLP framework for simultaneous model selection and identification is presented. The key issues for the successful design of this combined approach are: (i) selection of the integer and binary parameters that accurately describe all the possible nested models; (ii) reliable and accurate parameter estimation; (iii) use of efficient algorithms with reduced computational cost; (iv) assessment of model identifiability.
Nestedmodels: selection and identification
where p: set of parameters to be estimated NE: number of experiments N V_{ i }: number of measured variables in experiment i N M_{ ij }: number of measurements of the variable j in experiment i${\sigma}_{\mathit{\text{ijk}}}^{2}$: variance of the measurement k of the variable j in experiment i${\stackrel{~}{y}}_{\mathit{\text{ijk}}}$: measurement k of the variable j in experiment i y_{ ijk }: model predicted value k of the variable j in experiment i
Many functions have been suggested to compare two or more models. Despite the fact that several authors have questioned whether AIC is biased towards complex model structures [26], this function has been widely used as a metric to select the most adequate among hierarchical or nested models, since it encompasses model performance and model complexity [27]. It allows establishing a ranking of the models where the most adequate is the one with the smallest value of the criterion [25]. However, most of the available techniques for model selection based on the AIC, require the previous fitting of all the candidate models. Therefore, when the number of possible models is large or the simulation of the models is computationally expensive, these methodologies can become practically impossible [27].
In order to reduce the computational burden, in this work we used the AIC as cost function for finding the optimal set of parameters formed by a subset of binary parameters defining the model structure (e.g. presence or absence of a certain feedback loop) and another subset of integer and real parameters characterizing the model dynamics.
Formulation of the MINLP
subject to:

System dynamics in the form of DAEs, with state variables y$\begin{array}{c}f(\stackrel{\u0307}{\mathbf{y}},\mathbf{y},\mathbf{p},\mathbf{q})=0\phantom{\rule{2em}{0ex}}\hfill \end{array}$(4)$\begin{array}{c}\phantom{\rule{3em}{0ex}}\mathbf{y}\left({t}_{0}\right)={y}_{0}\phantom{\rule{2em}{0ex}}\hfill \end{array}$(5)

Additional requirements in the form of equality and/or inequality constraints$\begin{array}{l}h(\mathbf{y},\mathbf{p},\mathbf{q})=0\end{array}$(6)$\begin{array}{l}g(\mathbf{y},\mathbf{p},\mathbf{q})\le 0\end{array}$(7)

Upper and lower bounds (superscripts U and L respectively) on decision variables$\begin{array}{l}{p}^{L}\le \mathbf{p}\le {p}^{U}\end{array}$(8)$\begin{array}{l}{q}^{L}\le \mathbf{q}\le {q}^{U}\end{array}$(9)
This set of constraints defines the feasible space S, while the feasible objective space o is the set J (p, q)  (p, q) ∈ S.
Solution of the MINLP problem
The problem of parameter estimation is a crucial step in the development of models of biological systems [17]. Due to the nonlinear and dynamic nature of these systems and the usually sparse and noisy nature of the experimental data available, the resulting parameter estimation problem is frequently illconditioned and multimodal. Therefore, traditional local methods may fail and there is a need to use more sophisticated techniques as global optimization (GO) to successfully fit the model. In our group, different efforts have been devoted to design metaheuristics for efficient and robust parameter estimation in biological models [28, 29].
In the case of the MINLP problem at hand, the need to use GO methods is increased by the additional nonlinearities coming from the binary and integer variables and the augmented size of the problem. ACOmi (Ant Colony Optization for mixedinteger problems) [30] and fSSm [31] are robust extensions of metaheuristics (Ant Colony optimization and Scatter Search, respectively) that enable the handling of mixedinteger variable search domains; therefore, they are ideal for solving the MINLP problem introduced in this work.
ACOmi
ACOmi (Ant Colony Optization for mixed integer problems) is an extension of the ant colony optimization metaheuristic that enables to handle mixed integer variable search domains. In this method a new penalization strategy was introduced in order to extend the ACO framework to face constrained optimization problems. A detailed explanation of the hybrid implementation ACOmi, incorporating the extended ACO framework and a robust oracle penalty method, is given by [30].
fSSm
fSSm is a new evolutionary method for complexprocess optimization. It is partially based on the principles of the scatter search methodology, but making use of innovative strategies to be more effective in the context of complexprocess optimization using a small number of tuning parameters. In particular, this method uses a new combination method based on path relinking, which considers a broader area around the population members than previous combination methods. It also uses a populationupdate method which improves the balance between intensification and diversification, as described in [31]. Although fSSm is mainly designed for continuous problems, a rounding operator has been implemented for handling integer and binary variables.
MISQP
MISQP is a modified sequential quadratic programming method for solving MINLP problems. MISQP assumes that the model functions are smooth in the sense that an increment of an integer variable by one leads to a small change of function values but it does not require that the mixedinteger program is convex or relaxable (i.e. function values are evaluated only at integer points). Thus, this algorithm is expected to be more efficient than any other method that starts from a solution of the relaxed problem [32].
Moreover, in contrast to other local optimization solvers, the evaluation of the exact gradient is not always required for a proper convergence of SQP methods. The evaluation of the performance of the method used in this study, MISQP, on a test set of 186 academic test examples published in [33] showed that analytical partial derivatives subject to the integer variables do not improve robustness or efficiency, and the number of iterations is enlarged [34]. Diehl et al. [35] presented another SQP algorithm which does not require the evaluation of the exact constraint Jacobian matrix.
Model identifiability, sensitivity and correlation analysis
If the FIM is full rank the parameters are considered identifiable [42]. The parameters of a model are not identifiable when an infinite number of parameter sets fitting the experimental data with the same accuracy exist and the confidence intervals are infinite. Moreover, it is also interesting to study the parameter sensitivity and the correlation among parameters.
Sensitivity analysis
A high value of the sensitivity index means that a change in parameter p_{ θ } has an important effect on the model outcome making the parameter p_{ θ } identifiable with the data available if all the other parameters are fixed. Unless a parameter is unidentifiable due to total correlation with another parameter, the higher the sensitivity the more accurately the parameter can be identified and, on the other hand, a parameter with a small sensitivity will be very difficult to identify since any change on its value will have almost no influence on the model dynamics. Therefore, values of critical parameters should be thoroughly identified while parameters having a little effect can be simplified or even ignored [44].
The main drawback of local sensitivity indices is that they are computed at the nominal values used for the parameters and the behavior of the response function is described only locally in the parameter space. Moreover, preliminary experiments and parameter estimation tests should be carried out in order to obtain a first guess for the parameter values and an iterative scheme involving both steps is required to study the model sensitivity. In addition, these methods are linear; thus, they are not sufficient for dealing with complex models, especially those in which there are nonlinear interactions between parameters.
In contrast, global sensitivity analysis (GSA) methods evaluate the effect of a parameter while all other parameters are varied simultaneously, accounting for interactions between parameters without depending on the stipulation of a nominal point. In this work, a pseudoglobal sensitivity analysis as described in [39] was used. For that, 2^{10} sampling points were generated in the parameter space by means of Sobol’ lowdiscrepancy sequences that guarantee an uniform distribution avoiding clustering and empty areas [45]. Then, Bayesian Derivative based Global Sensitivity Measures [39] were computed using SensSB toolbox [46] and their metrics were used to establish an importance ranking for the parameters.
Correlation analysis
For models with several parameters, high parameter sensitivity, although necessary, does not ensure the identifiability of the model. In addition, the sensitivity functions of the parameters have to be linearly independent so a change in one parameter can not be compensated by changes in the other parameters. When the parameters are identifiable, we can study the degree of linear dependence among the sensitivity functions by means of a correlation analysis based on the Fisher Information Matrix (FIM) as described in [28]. This method requires the inversion of the FIM so it can only be applied when the parameters are identifiable and the FIM full rank. However, correlations among parameters close to +1 or 1 mean that the parameters are difficult to identify and the confidence intervals very large (although not infinite as in the case of nonidentifiable parameters). In that case, the model should be reduced by fixing some of the parameters to their nominal values or by properly grouping some sets.
In order to eliminate the dependence on a nominal point, a pseudoglobal identifiability analysis as described in [39, 46] was used. A correlation matrix was computed for each set of parameters used for the sensitivity analysis and a weighted average was obtained based on the maximum likelihood of each of the parameter sets. In this way, the influence of parameters notlikely to fit the data is minimized while the dependence on a nominal point is avoided.
Dynamic model of the KdpD/KdpE system of E scherichia coli
Bacteria constantly monitor their environment and adapt immediately to changing conditions to survive. There are several adaptation mechanisms notably special signal transduction systems. A sensor kinase (KdpD) and a response regulator (KdpE) regulate the expression of the KdpFABC operon encoding the high affinity K^{+} uptake system of Escherichia coli. In [47], a mathematical model for the KdpD/KdpE twocomponent system was developed and calibrated with the available in vitro and in vivo experimental data. The model can be separated into two submodels connected in a unidirectional way. The parameters corresponding to the signal transduction part were estimated from in vitro data while the parameters of the gene expression functional unit were identified from mRNA and KdpFABC concentrations determined in vivo using various extracellular stimulus, $S=\frac{{K}^{+}}{{K}_{0}^{+}}$.
where mRNA represents the concentration of messenger RNA, K d p D_{0} the total concentration of the sensor kinase, K d p D^{ P } the concentration of the phosphorylated KdpD, K d p E_{0} the total concentration of the response regulator, K d p E^{ P } the concentration of the phosphorylated KdpE, KdpFABC the concentration of the protein complex and ${\mathit{\text{KdpE}}}_{f}^{P}$ the concentration of the unbound response regulator.
Results and discussion
Computations were carried out using Matlab™ (Version 7.9.0, R2009b; The Mathworks, MA, USA) running on a dual INTEL®;XEON®;2.13 GHz CPU desktop under Windows 7. All the scripts needed to reproduce the results presented in the following are provided in the Additional file 1.
Identifiability analysis of the original model
Simulation studies showed that the concentration of K d p D^{ P } was very low and almost in steady state. Therefore, we removed equation (15) from the model and consequently we fixed the concentration of K d p D^{ P } to its initial value and parameters k_{1} and k_{1} were eliminated.
A local identifiability analysis of the original model with the best set of parameters was performed. As already suggested by Kremling et al[47], the full set of parameters is not uniquely identifiable with the available in vivo data; thus, some of the parameters were fixed to literature values or to values obtained from in vitro data.
The importance ranking of the parameters estimated from the in vivo data revealed that parameter k_{2} has the lowest sensitivity index (it accounts for the 0.002% of the total model sensitivity) while the two most relevant parameters, k_{ tr } and D N A_{0} represent 75% of the total sensitivity. Hence, k_{2} was fixed to its nominal value and special attention was payed to the set of most influential parameters. The parameter μ presented high correlations with other parameters so it was fixed to its experimental value μ = 0.5 l/h. Other pairs of parameters showed also high correlation among them but they could still be identified.
These modifications led to a second formulation of the model (Model II) with 7 DAEs and 17 parameters that fits the experimental data equally well.
New hypotheses for the KdpD/KdpE two component system
where

Regulation of translation (R_{1})${R}_{1}=\left\{\begin{array}{cc}1,\hfill & \text{if}\phantom{\rule{1em}{0ex}}{\mathit{\text{bin}}}_{1}=0\\ \frac{1}{\mathit{\text{KdpFAB}}{C}^{{n}_{1}}\phantom{\rule{0.3em}{0ex}}+\phantom{\rule{0.3em}{0ex}}{k}_{\mathit{\text{trans}}}},& \text{if}\phantom{\rule{1em}{0ex}}{\mathit{\text{bin}}}_{1}=1\end{array}\right.$(22)

Regulation of proteolysis (R_{2})${R}_{2}=\left\{\begin{array}{cc}1,\hfill & \text{if}\phantom{\rule{1em}{0ex}}{\mathit{\text{bin}}}_{2}=0\\ \frac{\mathit{\text{KdpFAB}}{C}^{{n}_{2}}}{\mathit{\text{KdpFAB}}{C}^{{n}_{3}}\phantom{\rule{0.3em}{0ex}}+\phantom{\rule{0.3em}{0ex}}{k}_{\mathit{\text{deg}}}},& \text{if}\phantom{\rule{1em}{0ex}}{\mathit{\text{bin}}}_{2}=1\end{array}\right.$(23)

Stimulus counteraction (R_{3})
Note that the dynamics of the mutant strain do not depend on parameters b i n_{3}, n_{4}, n_{5}, k_{ hy }, and k_{ hy }.
These possible new loops were integrated with the Model II considering a superstructure, which has a total of 25 degrees of freedom: 17 reals, 5 integers and 3 binaries, resulting in 1700 nested models. In a traditional setting, each of this model should be identified (calibrated) from experimental data by solving the corresponding minimization problem, that is, a nonlinearprogramming problem subject to differentialalgebraic constraints (NLPDAEs), prior model selection. Since the solution of each problem is quite computationally expensive, this is obviously not tractable. As an alternative, we applied the strategy outlined above and performed a simultaneous selection and identification via MINLP optimization.
MINLP solutions
In order to illustrate the capabilities of the methodology presented in this work, we generated in silico data via simulation using a nominal set for parameters and a certain model structure. Starting from a known structure and known parameter values allows us to asses the performance of the MINLP formulation by checking if it is able to recover the original model.
Nominal value for the parameters and MINLP best solution
Parameter  Nominal value  MINLP solution 

k _{2}  5.18E+07  4.74E+07 
k _{3}  9.76E+01  1.20E+02 
α  5.79E02  5.23E02 
k _{ tr }  1.00E+03  8.66E+02 
k _{ tl }  4.96E+03  4.28E+03 
k _{t l 2}  1.03E+03  1.27E+03 
k _{t l 3}  2.05E+03  1.64E+03 
k _{ z }  4.99E+01  6.06E+01 
k _{d 2}  1.00E+01  1.25E+01 
D N A _{0}  6.16E04  7.08E04 
K _{ a }  1.82E07  2.11E07 
K  1.00E+03  8.48E+02 
k _{ d }  1.18E+00  9.55E01 
k _{ hy }  2.00E+06  1.76E+06 
k _{ transf }  9.74E01  8.07E01 
k _{ degf }  1.36E01  1.84E01 
k _{ hy }     
n_{1}:  3  3 
n_{2}:  1  1 
n_{3}:  2  3 
n_{4}:  3  3 
n_{5}:  0  0 
b i n_{1}:  1  1 
b i n_{2}:  1  1 
b i n_{3}:  0  0 
Checking the multimodality of the MINLP
Identifiability analysis of the resulting model
The FIM computed for the best set of parameters obtained by the global solver is full rank; therefore, we can assert that the parameters are locally identifiable.
The correlation matrix shows several pairs of parameters highly correlated what explains the difficulties encountered by the local method in finding the global solution. Despite the identifiability difficulties of this problem, which make most of the solvers fail when trying to solve it, the residuals for the solution obtained by fSSm are small indicating a precise parameter estimation, e.g., the estimated values are close to the experimental data.
Methodology strengths and limitations
The goal of this study is not to solve the general problem of model inference but a dense subcase of it, i.e., the discrimination among a subset of nested competing models and simultaneous estimation of the model parameters. In other words, we consider the very frequent situation in systems biology where a first model is available based on previous knowledge but new experimental information allow to formulate different hypotheses to refine it. Thus, instead of solving a general inference problem (i.e. find the model structure plus the parameters from a set of data), we consider a subproblem which is smaller (although still very challenging) and dense (so sparsity is not an issue), and which, therefore, does not suffer from many of the illposedness and illconditioning maladies of the general inference problem [51, 52]. Despite the usefulness and broad applicability of the presented approach for model development in systems biology, there are three major limitations worth mentioning here:

Scaling up to largescale models: the corresponding MINLPs might become rather large and therefore the computational effort needed for their solution might become prohibitive.

Nonuniqueness of biochemical reaction mechanisms: it is known that biochemical reaction networks with different structure and/or parametrization may produce the same dynamic response describing the timeevolution of species concentrations (see the recent discussion and results in [52]) difficulting the solution of the associated MINLPs. Fortunately, and following our comments above, this work considers a dense subcase of the general inference problem, so these issues are not as important. In fact, our approach can be interpreted as the application of extra constraints that can be used to reduce uniqueness and identifiability issues.

Model identification/selection metric: the use of more advanced metrics for model selection such as the likelihood ratio or the Ftest can not be used in this approach since they rely on pairwise comparisons. However, in the presented methodology the AIC could be replaced by any other metric for model selection as long as it can establish a ranking for the set of competing models encompassing model performance and model complexity.
Conclusions
We consider this cycle in a more compact way grouping sets of hypotheses together and formulating a parameterized set of nested models, from which the best alternative must be selected. We then formulate the decision problem as an MINLPbased optimization for simultaneous model selection and parameter identification.
This procedure has been applied to a case study considering potassium homeostasis in bacteria, arriving to the following conclusions: (i) the presented MINLPbased approach for nestedmodel selection is a powerful methodology for model selection and identification in systems biology; and (ii) for the case study considered, it has resulted in a model that presents a better fit to the in silico generated experimental data.
Declarations
Acknowledgements
Authors MRF and JRB acknowledge financial support from the EU ERASysBio programme and the Spanish MICINN and MINECO (SYSMO grant KOSMOBAC, ref. GEN200627747E/SYS and project MultiScales ref. DPI201128112C0403, both with partial support from the European Regional Development Fund, ERDF). MR was supported by the Max Planck society and the European Erasmus project. AK was funded in part by the BMBF through the EraNet initiative SysMO. We acknowledge support of the publication fee by the CSIC Open Access Publication Support Initiative through its Unit of Information Resources for Research (URICI).
Authors’ Affiliations
References
 van Riel NAW: Dynamic modelling and analysis of biochemical networks: mechanismbased models and modelbased experiments. Brief Bioinform. 2006, 7 (4): 36410.1093/bib/bbl040.PubMedView Article
 Stelling J: Mathematical models in microbial systems biology. Curr Opin Microbiol. 2004, 7 (5): 513518. 10.1016/j.mib.2004.08.004.PubMedView Article
 Banga JR, BalsaCanto E: Parameter estimation and optimal experimental design. Essays Biochem. 2008, 45: 19510.1042/BSE0450195.PubMedView Article
 Jaqaman K, Danuser G: Linking data to models: data regression. Nat Rev Mol Cell Biol. 2006, 7 (11): 813819. 10.1038/nrm2030.PubMedView Article
 Ashyraliyev M, FomekongNanfack Y, Kaandorp JA, Blom JG: Systems biology: parameter estimation for biochemical models. FEBS J. 2008, 276 (4): 886902.View Article
 Kitano H: Computational systems biology. Nature. 2002, 420 (6912): 206210. 10.1038/nature01254.PubMedView Article
 Aderem A: Systems biology: its practice and challenges. Cell. 2005, 121 (4): 511513. 10.1016/j.cell.2005.04.020.PubMedView Article
 Arita M, Robert M, Tomita M: All systems go: launching cell simulation fueled by integrated experimental biology data. Curr Opin Biotechnol. 2005, 16 (3): 344349. 10.1016/j.copbio.2005.04.004.PubMedView Article
 Feng X, Rabitz H: Optimal identification of biochemical reaction networks. Biophys J. 2004, 86 (3): 12701281. 10.1016/S00063495(04)742010.PubMedPubMed CentralView Article
 Kremling A, Fischer S, Gadkar K, Doyle III FJ, Sauter T, Bullinger E, Gilles ED, Allgower F: A benchmark for methods in reverse engineering and model discrimination: problem formulation and solutions. Genome Res. 2004, 14 (9): 177310.1101/gr.1226004.PubMedPubMed CentralView Article
 Gadkar KG, Gunawan R, Doyle III FJ: Iterative approach to model identification of biological networks. BMC Bioinform. 2005, 6: 15510.1186/147121056155.View Article
 BalsaCanto E, Alonso A, Banga JR: An iterative identification procedure for dynamic modeling of biochemical networks. BMC Syst Biol. 2010, 4: 1110.1186/17520509411.PubMedPubMed CentralView Article
 Bandara S, Schloeder JP, Eils R, Bock HG, Meyer T: Optimal experimental design for parameter estimation of a cell signaling model. PLoS Comput Biol. 2009, 5 (11): e100055810.1371/journal.pcbi.1000558.PubMedPubMed CentralView Article
 Apgar JF, Toettcher JE, Endy D, White FM, Tidor B: Stimulus design for model selection and validation in cell signaling. PLoS Comput Biol. 2008, 4 (2): e3010.1371/journal.pcbi.0040030.PubMedPubMed CentralView Article
 Melykuti B, August E, Papachristodoulou A, ElSamad H: Discriminating between rival biochemical network models: three approaches to optimal experiment design. BMC Syst Biol. 2010, 4: 3810.1186/17520509438.PubMedPubMed CentralView Article
 Skanda D, Lebiedz D: An optimal experimental design approach to model discrimination in dynamic biochemical systems. Bioinform. 2010, 26 (7): 939945. 10.1093/bioinformatics/btq074.View Article
 Lillacci G, Khammash M: Parameter estimation and model selection in computational biology. PLoS Comput Biol. 2010, 6 (3): e100069610.1371/journal.pcbi.1000696.PubMedPubMed CentralView Article
 Verheijen PJT: Model selection: an overview of practices in chemical engineering. Comput Aided Chem Eng. 2003, 16: 85104.View Article
 Petzold L, Zhu W: Model reduction for chemical kinetics: An optimization approach. AIChE J. 1999, 45 (4): 869886. 10.1002/aic.690450418.View Article
 Edwards K, Edgar TF, Manousiouthakis VI: Reaction mechanism simplification using mixedinteger nonlinear programming. Comput Chem Eng. 2000, 24: 6779. 10.1016/S00981354(00)003112.View Article
 Maurya MR, Bornheimer SJ, Venkatasubramanian V, Subramaniam S: Mixedinteger nonlinear optimisation approach to coarsegraining biochemical networks. IET Syst Biol. 2009, 3: 2410.1049/ietsyb:20080098.PubMedPubMed CentralView Article
 Nikolaev EV: The elucidation of metabolic pathways and their improvements using stable optimization of largescale kinetic models of cellular systems. Metab Eng. 2010, 12 (1): 2638. 10.1016/j.ymben.2009.08.010.PubMedView Article
 Banga JR: Optimization in computational systems biology. BMC Syst Biol. 2008, 2: 4710.1186/17520509247.PubMedPubMed CentralView Article
 Aldrich J: R. A. Fisher and the making of maximum likelihood 19121922. Stat Sci. 1997, 12 (3): 162176.View Article
 Akaike H: A new look at the statistical model identification. IEEE Trans Automatic Control. 1974, 19 (6): 716723. 10.1109/TAC.1974.1100705.View Article
 Cedersund G, Roll J: Systems biology: model based evaluation and comparison of potential explanations for given biological data. FEBS J. 2009, 276 (4): 903922. 10.1111/j.17424658.2008.06845.x.PubMedView Article
 McDonald CP, Urban NR: Using a model selection criterion to identify appropriate complexity in aquatic biogeochemical models. Ecol Model. 2010, 221 (3): 428432. 10.1016/j.ecolmodel.2009.10.021.View Article
 RodriguezFernandez M, Mendes P, Banga JR: A hybrid approach for efficient and robust parameter estimation in biochemical pathways. Biosyst. 2006, 83 (23): 248265. 10.1016/j.biosystems.2005.06.016.View Article
 Egea JA, RodriguezFernandez M, Banga JR, Martí R: Scatter Search for Chemical and BioProcess Optimization. J Glob Optimization. 2007, 37: 481503. 10.1007/s1089800690753.View Article
 Schlüter M, Egea JA, Banga JR: Extended ant colony optimization for nonconvex mixed integer nonlinear programming. Comput Oper Res. 2009, 36 (7): 22172229. 10.1016/j.cor.2008.08.015.View Article
 Egea JA, Banga JR, Martí R: An evolutionary method for complexprocess optimization. Comput Oper Res. 2010, 37 (2): 315324. 10.1016/j.cor.2009.05.003.View Article
 Exler O, Schittkowski K: A trust region SQP algorithm for mixedinteger nonlinear programming. Optimization Lett. 2007, 1 (3): 269280. 10.1007/s1159000600261.View Article
 Schittkowski K: A collection of 186 test problems for nonlinear mixedinteger programming. 2012, Tech. rep., Department of Computer Science; University of Bayreuth
 Exler O, Lehmann T, Schittkowski K: A comparative study of SQPtype algorithms for nonlinear and nonconvex mixedinteger optimization. Math Program Comput. 2012, 4 (4): 383412. 10.1007/s1253201200450.View Article
 Diehl M, Walther A, Bock HG, Kostina E: An adjointbased SQP algorithm with quasiNewton Jacobian updates for inequality constrained optimization. Optimization Methods Softw. 2010, 25 (4): 531552. 10.1080/10556780903027500.View Article
 Raue A, Kreutz C, Maiwald T, Bachmann J, Schilling M, Klingmueller U, Timmer J: Structural and practical identifiability analysis of partially observed dynamical models by exploiting the profile likelihood. Bioinform. 2009, 25 (15): 19231929. 10.1093/bioinformatics/btp358.View Article
 Vanlier J, Tiemann CA, Hilbers PAJ, van Riel NAW: An integrated strategy for prediction uncertainty analysis. Bioinform. 2012, 28 (8): 11301135. 10.1093/bioinformatics/bts088.View Article
 Brannmark C, Palmer R, Glad ST, Cedersund G, Stralfors P: Mass and information feedbacks through receptor endocytosis govern insulin signaling as revealed using a parameterfree modeling framework. J Biol Chem. 2010, 285 (26): 2017120179. 10.1074/jbc.M110.106849.PubMedPubMed CentralView Article
 RodriguezFernandez M, Banga JR, Doyle III FJ: Novel global sensitivity analysis methodology accounting for the crucial role of the distribution of input parameters: application to systems biology models. Int J Robust Nonlinear Control. 2012, 22 (10): 10821102. 10.1002/rnc.2797.View Article
 Jacquez JA, Greif P: Numerical Parameter Identifiability and estimability: integrating identifiability, estimability, and optimal sampling desing. Math Biosci. 1985, 77: 201227. 10.1016/00255564(85)900987.View Article
 Audoly S, Bellu G, D’Angio L, Saccomani MP, Cobelli C: Global identifiability of nonlinear models of biological systems. IEEE Trans Biomed Eng. 2001, 48 (1): 5565. 10.1109/10.900248.PubMedView Article
 Fisher F: Econometrica. 1959, 27 (3): 431447. 10.2307/1909470.View Article
 Brun R, Reichert P, Kunsch HR: Practical identifiability analysis of large environmental simulation models. Water Resour Res. 2001, 37: 10151030. 10.1029/2000WR900350.View Article
 Karnavas WJ, Sanchez P, Bahill AT: Sensitivity analyses of continuous and discrete systems in the time and frequency domains. IEEE Trans Syst, Man, Cybern. 1993, 23 (2): 488501. 10.1109/21.229461.View Article
 Kucherenko S, RodriguezFernandez M, Pantelides C, Shah N: Monte Carlo evaluation of derivative based global sensitivity measures. Reliability Eng Syst Saf. 2009, 94: 11351148. 10.1016/j.ress.2008.05.006.View Article
 RodriguezFernandez M, Banga JR: SensSB: A software toolbox for the development and sensitivity analysis of systems biology models. Bioinform. 2010, 26 (13): 16751676. 10.1093/bioinformatics/btq242.View Article
 Kremling A, Heermann R, Centler F, Jung K, Gilles ED: Analysis of twocomponent signal transduction by mathematical modeling using the KdpD/KdpE system of Escherichia coli. Biosyst. 2004, 78 (13): 2337. 10.1016/j.biosystems.2004.06.003.View Article
 Alon U: An Introduction to Systems Biology  Design Principles of Biological Circuits. 2007, London: Chapman & Hall/CRC
 Widder S, Schicho J, Schuster P: Dynamic patterns of gene regulation I: Simple twogene systems. J Theor Biol. 2007, 246 (3): 395419. 10.1016/j.jtbi.2007.01.004.PubMedView Article
 Konkoli Z: A danger of low copy numbers for inferring incorrect cooperativity degree. Theor Biol Med Model. 2010, 7: 4010.1186/17424682740.PubMedPubMed CentralView Article
 Prill R, Marbach D, SaezRodriguez J, Sorger P, Alexopoulos L, Xue X, Clarke N, AltanBonnet G, Stolovitzky G: Towards a rigorous assessment of systems biology models: the DREAM3 challenges. PloS one. 2010, 5 (2): e920210.1371/journal.pone.0009202.PubMedPubMed CentralView Article
 Szederkenyi G, Banga JR, Alonso AA: Inference of complex biological networks: distinguishability issues and optimizationbased solutions. BMC Syst Biol. 2011, 5: 17710.1186/175205095177.PubMedPubMed CentralView Article
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.