Volume 10 Supplement 12
Bioinformatics Methods for Biomedical Complex System Applications
Developing optimal input design strategies in cancer systems biology with applications to microfluidic device engineering
 Filippo Menolascina^{1, 2, 5}Email author,
 Domenico Bellomo^{3, 6, 7},
 Thomas Maiwald^{4},
 Vitoantonio Bevilacqua^{1},
 Caterina Ciminelli^{1},
 Angelo Paradiso^{5} and
 Stefania Tommasi^{5}
DOI: 10.1186/1471210510S12S4
© Menolascina et al; licensee BioMed Central Ltd. 2009
Published: 15 October 2009
Abstract
Background
Mechanistic models are becoming more and more popular in Systems Biology; identification and control of models underlying biochemical pathways of interest in oncology is a primary goal in this field. Unfortunately the scarce availability of data still limits our understanding of the intrinsic characteristics of complex pathologies like cancer: acquiring information for a system understanding of complex reaction networks is time consuming and expensive. Stimulus response experiments (SRE) have been used to gain a deeper insight into the details of biochemical mechanisms underlying cell life and functioning. Optimisation of the input timeprofile, however, still remains a major area of research due to the complexity of the problem and its relevance for the task of information retrieval in systems biologyrelated experiments.
Results
We have addressed the problem of quantifying the information associated to an experiment using the Fisher Information Matrix and we have proposed an optimal experimental design strategy based on evolutionary algorithm to cope with the problem of information gathering in Systems Biology. On the basis of the theoretical results obtained in the field of control systems theory, we have studied the dynamical properties of the signals to be used in cell stimulation. The results of this study have been used to develop a microfluidic device for the automation of the process of cell stimulation for system identification.
Conclusion
We have applied the proposed approach to the Epidermal Growth Factor Receptor pathway and we observed that it minimises the amount of parametric uncertainty associated to the identified model. A statistical framework based on MonteCarlo estimations of the uncertainty ellipsoid confirmed the superiority of optimally designed experiments over canonical inputs. The proposed approach can be easily extended to multiobjective formulations that can also take advantage of identifiability analysis. Moreover, the availability of fully automated microfluidic platforms explicitly developed for the task of biochemical model identification will hopefully reduce the effects of the 'data richdata poor' paradox in Systems Biology.
Background
Our understanding of molecular basis of complex diseases is being dramatically changed by systems investigation supported by the most advanced tools and techniques developed by the scientific community. In particular, cancer investigation has greatly benefited by systems level approaches since tumor development and progression are believed to be among those system trajectories that arise from abnormal working states. The work by Hornberg and colleagues [1] pointed out the relevance of Systems Biology approaches in the study of dynamics leading to cancer. Epidermal Growth Factor Receptor (EGFR) pathway is one of those biochemical reaction networks believed to play a central role in cancer development. As a matter of fact EGFR and receptors in the same family (ErbB2, ErbB3 and ErbB4) mediate cell to cell interactions both in organogenesis and in adult tissues [2]. The 40year long study of this pathway led to associate overexpression of the EGFR family members to several types of cancer [3]. Because of the high clinical relevance, several efforts have been spent in the last decades in unravelling the complex dynamics of this biochemical network, as well as in finding potential targets of therapeutic intervention [4–6]. Although global models of EGFR pathway exist [7–12], many questions still remain open both in terms of model accuracy [13–15], parameter identifiability [16] and driving input design [17, 18]. In this context we put the pioneering works by Arkin and colleagues [19–22], van Oudenaarden and colleagues [23] and Steuer and colleagues [24]. Other recent works have focused on the connections between optimal experimental design strategies and structural and experimental identifiability analysis of biochemical pathways; this is the case of [16, 25–28].
Structural identifiability refers to the possibility of finding the mathematical model of the true system (see [29, 30] for references in biological systems investigation), after having applied a specific search strategy in the space of the solutions.
Experimental identifiability [31], on the other hand, is related to the possibility of finding the mathematical representation of the true model given a predetermined set of observations. This is a central aspect of this class of identifiability problems since it is more focused on the available data and, in particular, on information content. This aspect establishes an interesting bridge between System Identification Theory and Experimental Design. The Design of Experiments (DOE) is a well developed methodology in statistics [32] focusing on the design of all informationgathering exercises where variation is present, the main objective of the whole task being the maximisation of the information obtained from experiment and the minimisation of the number of experiments. This specific task is commonly referred to as 'Optimal Experimental Design' (OED). This discipline quickly gained a significant interest among researchers mostly in natural and social sciences but became an active research field in engineering only with the pioneering work by Lennart Ljung and his standard model for dynamical system identification oriented experiments [33]. This model has been recently modified by Phair et al. [34] and Cho et al. [35].
Nevertheless the main idea behind system identification in Systems Biology remained unchanged [36]. In line with Fisher's criteria, Ljung's scheme [33] suggests to define a detailed plan of the experiments to be carried out before starting to collect inputoutput data from the system to be modeled. Specifications like data sampling strategies and driving inputs should be fixed in order to optimise the information yield of each experiment and to address the cost minimisation task OED is aimed at. These issues gain an even stronger relevance, if we consider the socalled 'data richdata poor paradox' [37] resulting from the difficulties and costs involved in Systems Biology related assays. For these reasons and in order to develop a comprehensive framework for system identification in Systems Biology, we will describe how a specific issue of OED, namely Optimal Input Design (OID), can be addressed using optimality criteria and microfluidicsbased experiments. As a matter of fact, microfluidic platforms have been shown to provide a powerful tool for the development of datarich experimental strategies able to fill the gap of the previously cited paradox. Signals obtained in this stage are used as templates for the development of a microfluidic device for a flexible and automated platform for affordable singlecell experiments in Systems Biology.
In the following paragraphs, we go through a brief introduction of the EGFR model then we analyse OED and OID criteria. We review current approaches to cell stimulation in the 'Methods' section and compare them with optimality criteria derived ones. An analysis of the experimental results follows in 'Results', where we introduce a feasible design of the microfluidic device thought to speed up the process of data collection in Systems Biology by lowering the costs associated to experiments. A discussion of the results presented herein and final cues for further research are given in the last paragraph.
Results and discussion
In order to model and understand the functionality of the EGFR signaling cascade a quantitative description of the signal dynamics is of major relevance. For this reason, we discuss the computational results obtained from insilico simulations carried out using POTTERS WHEEL[38]. POTTERWHEEL is a multiexperimental fitting MATLAB package intended to allow researchers to ease model analysis and experimental setup steps. In particular, this package is one of the few in Systems Biology providing a simple interface to external input based simulation of biological pathways behavior.
To estimate the effects of different inputs on parameters estimates uncertainties, we carried out 1000 identification experiments for each of the three classes of stimuli, namely: a step input, a persistently exciting input and the time profile of the stimulation obtained from the optimisation task. Therefore we plotted a bivariate distribution of both the V_{ max }and K_{ m }of the first MichaelisMenten based reaction ( where v_{0} is the initial reaction rate and [S] the substrate concentration, see 'Methods' for a detailed description of the mathematical modelling step) in Kholodenko's model accounting for the dephosphorylation of the EGFEGFR dimer. It should be noted that parameter correlation can greatly affect our ability to successfully recover real parameters. This is one of the main issues arising in the field of parameter idenfitiability. In particular parameters that are structurally correlated cannot be uniquely identified from experiments. In order to investigate such peculiarities of our dynamical model we carried out an identifiability analysis based on the 'Alternating Conditional Expectation' algorithm (ACE) method described in [16] and implemented in the Mean Optimal Transformation Approach (MOTA) package. We present herein the computational results so as to provide a tool for comparing the different approaches to OED in Systems Biology.
Identifiability analysis
The statistical investigation of the properties of OED should be a primary goal when the time profile of the input is computed. Previous works in this field have focused on the comparison among alternative designs on the FIM cost values and confidence intervals [39]. Nevertheless it has been noticed that the FIM is derived from a linearisations of the least squares thus it may be unreliable in cases of considerably extended non linearities. The non identifiability of one of the parameters directly implies the functional relationships among at least two of them [16]. This phenomenon can be easily observed by plotting the joint probability distribution of each of them which will show statistically significant differences if compared with the expected multivariate normal one. From an algebraic point of view this results in the loss of rank of the covariance matrix of the multivariate distribution or, alternatively, a condition number of the same matrix asymtotically tending to infinity. Then, in order to cope with identifiability issues that may arise in identification tasks, we propose to investigate this property by using the ACE method proposed by Hengl and colleagues [16]. MOTA package is included in POTTERS WHEEL toolbox and is applied together with a linear fit sequence analysis. MOTA detects groups of two or more linearly or nonlinearly related parameters. It revealed some major nonidentifiabilities in the parameter space (results not shown) whose nature should be certainly investigated in order to understand their causes and possible solutions. We should remark that an integrated approach using MonteCarlo methods for both experimental design otpimisation and parameter correlation investigation can be a feasible choice. However, we should consider that this would imply a major rise in computational costs of the approach resulting fom the high number of parameter estimation tasks to be accomplished. Not to mention the issues arising from large scale models, noise and potential multimodality that would certainly imply using a robust global optimisation algorithm.
MonteCarlo based analysis

major semiaxis equal to λ_{ max }(Cov)

minor semiaxis equal to λ_{ min }(Cov)

rotational offset with respect to the the x axis equal to
Conclusion
The intrisic quantitative nature of Systems Biology poses new issues in everyday laboratory practice. Modelling, in this context, has long suffered from data shortage; the 'data richdata poor' paradox greatly influenced the pace towards a comprehensive understanding of molecular mechanisms governing biological systems. Nevertheless the potential of novel experimental techniques seems to promise new groundbreaking innovations thus increasing the versatility of new laboratory protocols and keeping experimentassociated costs low. Among these major limitations we should certainly mention the ability to stimulate cells in chemostats with input having very limited harmonic content. Microfluidic technology currently allows us to go beyond step like stimulation and to generate complex timevarying signals whose modulation can be achieved using control engineering strategies [42]. The availability of such tools and devices will allow us overcome the limits of indicial response for highly complex and fedback dynamical systems identification as outlined in [43, 44]. In this framework, the ability to optimally control and take advantage of the new methods and devices will be a major focus of the scientific community. This contribution presents, then, a mathematical formulation of the problem of optimal experimental design in Systems Biology by considering a case study of one of the most relevant biological pathways for cancer development. Formal derivation of problem definition results and heuristic solutions to a highly nonlinear optimisation problem have been both provided. In particular we formulated the problem of OED in Systems Biology as a nonlinear optimisation task in which the amount of information per experiment, quantified in the Fisher Information Matrix, is optimised by varying the stimulus time profile here representing the concentration of extracellular EGF ligand. We set up an evolutionary optimisation task aimed at finding the time sequence of the input signal that maximises the amount of information associated to the experiment. Moreover we proposed a statistical framework based on MonteCarlo estimates for the computation of the uncertainty regions for the parameter values; identifiability analysis, on the other hand, has been carried out using the ACE approach integrated in POTTERS WHEEL package. The results shown clearly indicate that dynamic experiments outperform canonical experiments based on sustained or persistently exciting inputs. Nonetheless we should consider that the approach presented herein depends on the starting model; a sequential experimental design should be investigated in order to overcome this issue. Moreover we should consider that all the simulations reported should be validated in a series of experiments. For this reason, we proposed the microfluidic device described in the 'Methods' section. 'Labrys' goes beyond the specific context of EGFR and, associated to a Hardware Abstraction Layer like Biosteram [45], is thought to provide researchers with a fully automated platform for complex experiment development and implementation. Future work in this field will certainly require a more tight collaboration among the different competences in the field of Systems Biology aimed at the full integration of both hardware and software findings for the development of a common, powerful and versatile platform for systems oriented experiments.
Methods
Model definition
In this work, we consider the EGFR signaling network model proposed by Kholodenko and colleagues in [7]. This model explores the short term pattern of cellular responses to epidermal growth factor (EGF) in isolated hepatocytes and predicts how the cellular response is controlled by the relative levels and activity states of signaling proteins and under what conditions activation patterns are transient or sustained. BioModels database [46] provides a selection of the most common file formats of this model. For our purposes, we will use the SBML version [47] featuring 25 molecular species, 23 reactions and 50 parameters. The set of ODEs describing this model can be extracted using COPASI [48] and translated in a MATLAB SIMULINK^{®} model for further simulations and analyses. Both massaction and MichaelisMenten kinetics have been used by Kholodenko and colleagues, resulting in a nonlinear model. In order to elicit this pathway with a driving input, we slightly modified it so as to include an external control for the ligand species, i.e. the Epidermal Growth Factor. For further experiments we selected POTTERS WHEEL[38] platform; this software provides highly powerful tools for the investigation of biological models' properties (just like parameter fitting and identifiability analysis) and, to the best of our knowledge, is one of the few ones in Systems Biology allowing researchers to easily define time evolution of forcing inputs without using complex formulations based on events and rules provided by SBML specifications.
We imported Kholodenko's model in POTTERS WHEEL and we edited the mfile so as to force EGF to be an input for our system and downstream species as our observables or outputs. This is a Single InputMulti Output (SIMO) formulation of the EGFR and will prove to be an interesting model for the stimulation of interesting behaviors (e.g. transmission blocking zeros elicitation which is a counterintuitive behavior of some dynamical systems which show a null output even if they are stimulated by nontrivial inputs with specific harmonic content). For our purpose, however, we will select only one downstream species to be observed just like in Single InputSingle Output systems whose study drove the field of dynamical system identification.
Dynamical systems identification
Time evolution of the system state x(t) ∈ ℝ^{ n }can be easily derived by solving the system of differential equations in Eq. 1 and imposing constraints on initial conditions x(0). Notice that the rate of change of x_{ i }depends, in general nonlinearly, on state variables x_{ j }, j = 1,..., n, on input trajectories u_{ i }and on parameters vector θ.
Notice that, in order to solve these ODEs, at least two quantities are required: k_{1} and the inititial concentration of A species, A(0). The parameters vector θ is usually intended to collect these quantities.
Accurate identification of parameters governing the dynamics of biochemical reaction networks is currently considered a major challenge in Systems Biology. In fact, even though some control on initial concentration of species can be obtained with experimental protocols (e.g. starvation), rate coefficients are driven by several external factors (temperature, PH etc.) in a very complex way. Moreover accurate parameters estimation is a key step for the elicitation of interesting behaviors in cellular pathways [49].
where we denoted the true parameter vector with θ_{0}. Here ε_{ i }∈ ℝ^{ p }describes the gaussian component of the error at time t_{ i }. We notice that the observation function g(·) together with the input function u(·) and the set of sampling times t fully defines the experimental design. In this we glimpse the triple nature of OED which aims at establishing optimal strategies for (a) sampling time [50], (b) species to be measured [17] and (c) input selection [18].
Optimal experimental design in systems biology
As previously stated OED has its main objective in maximising the information yield returned by an experiment. This is a central aspect in everyday practice in Systems Biology since experiments can be both highly expensive and timeconsuming, limiting practical fasibility of otherwise promising protocols. Applications of OED in Systems Biology have been described in [17, 51–55]; in particular [54, 56–60] have focused on model discrimination by OED. Experimental designs are usually categorised as starting and sequential designs.
In starting designs no data have been previously collected and the experimenter is interested in drawing the maximum amount of information from the experiment to be planned. This is done by minimising (or maximising) a specified objective function. Within this category we identify two subcategories: exact and continuous designs.
Exact designs have their own objective in the optimal placement of a finite number of design points [61]; on the other hand continuous designs deal with the selection of a design measure, η, which is equivalent to a probability density over the design space.
Sequential designs try to develop optimal strategies for model refinement of a preexisting model [62]. In this paper we will focus on a semisequential approach that can be considered a sequential design in that it starts from a compiled dynamic model of the EGFR pathway, but we do not use the results of this design to carry out further identification experiments since this would require non standard technological platforms. Given the potential impact of OED strategies on Systems Biology research, some researchers proposed software packages providing the user with significant opportunities for optimal experiments planning [63–65]. All of these packages are built on the principles of optimality and based on metrics being defined on the Fisher Information Matrix. This is a quite general framework for OED; unfortunately none of them currently provides a solution for Optimal Input Design. We will analyse this and other issues related to biochemical pathway stimulation in the next sections.
Optimal input design
Optimal Input Design for system identification provides several alternative measures of the identified model being used for optimal design [66]. Here we start by reporting some of the main results in the field [33] and then we discuss their implications in the specific case. Several contributions in this field focused on the minimisation of some measure of the variance of the estimated parameters like Fisher Information Matrix which can be used to estimate varianceuncertainty associated to parameter estimates. This process will be analysed in the next section; in this paragraph we will focus on a theoretical study of the OID for dynamic systems identification. Identification processes start with data collected on InputOutput behavior of the system under investigation.
Since Φ_{ z }(ω) is positive definite, this implies that almost everywhere that proves the previous statement. Moreover we can observe that, given the Φ_{ z }(ω), for the Schur's Lemma we can assure algorithm convergence only if Φ_{ u }(ω) > 0 and Φ_{ uu }(ω)  Φ_{ uy }(ω) Φ_{ yy }(ω) Φ_{ yu }(ω) > 0. Evidently the only block of this array we have control on is the one representing the spectrum of input signal which directly depends on dynamical properties of the driving input signal we design. It is therefore convenient to reintroduce the concept of persistently exciting signal of order n for a quasistationary stimulus u(t): we say that a similar signal, with spectrum Φ_{ u }(ω) is persistently exciting of order n is, for all filters of the form M_{ n }(q) = m_{1}q^{1} + ... + m_{ n }q^{ n }the relation implies that . Evidently the function M_{ n }(z) M_{ n }(z^{1}) can have n  1 different zeros on the unit circle (since one zero is always at the origin) taking symmetry into account. Hence u(t) is persistently exciting of order n if Φ_{ u }(ω) is different from zero on at least n points in the interval π ≤ ω ≤ π. This is a direct consequence of the definition.
Signals that show such properties have been investigated and include:

Pseudo Random Binary Signal

Generalised Binary Noise (or Random Binary Signal)

Sum of Sines and Filtered Noise

Coloured Noise
Notice that this is a rather general result for the class of systems considered herein. Nevertheless a similar argumentation can be carried out by considering an information metric directly tied to experimental data and to the model to be identified. These results are commonly derived from the analysis of some metric on the Fisher Information Matrix which are commonly referred to as 'Optimality criteria'.
Optimality criteria
we can easily derive a lower bound for the efficacy of a general and unbiased estimator that directly depends only on the Fisher Information Matrix. Evidently, the smaller the joint confidence intervals for the estimated parameters are, the more information the experiment carries with it. We can summarise the information about the variability in the covariance matrix into a single number by using metrics like Det(F), max(λ_{ i }) (with λ_{ i }representing the i^{ th }eigenvalue of F). This is where the alternative choice of optimality criteria arises. We can distinguish four major measures of the information content [32]:

AOptimal design: maximising trace(F)

DOptimal design: minimising Det(Σ)

EOptimal design: minimising λ_{ max }(Σ)

Modified EOptimal design: minimising
AOptimal designs are rarely used since they can lead to noninformative experiments [68]. DOptimal designs can be interpreted as geometric means minimisation of the errors in the parameter estimates. EOptimal and Modified EOptimal designs try to minimise the largest uncertainty and the ratio of the largest and smallest uncertainties among parameter estimates respectively. Given the characteristics of each of these criteria and the computational efforts required for the specific problem we selected DOptimality as driving criterion for our input design task.
Computational implementation
In order to carry out the optimisation of the input time profile we set up an optimisation routine based on a Genetic Algorithm (GA) thought to minimise an objective function encoding the DOptimality metric on the FIM. Here we present a pseudocode of the proposed approach.
Algorithm 1 Genetic_Algorithm(population, Fitness_Function) returns individual
while {an individual has a fitness value higher(lower)than threshold} do
new _population ← ∅
for i from 1 to Dim(population) do
x ← Random _Selection(population, Fitness _Function)
y ← Random _Selection(population, Fitness_Function)
descendant ← Reproduction(x, y)
if small _random _probability ← new _population then
new _population ← new _population ∪ descendant
end if
population ← new _population
end for
end while
return best individual evaluated on the Fitness _Funtion
Algorithm 2 Reproduction(x, y) returns individual
x ← Length(x)
c ← random number in the range {0, n}
return Append(Substring(x, 1, c), Substring(y, c + 1, n))
Algorithm 3 Fitness Function x(x) returns fitness value
input ← x
time _evolution ← Simulate(EGFR pathway, input, θ)
FIM ← Fisher _Information _Matrix(time _evolution)
return FIM
Algorithm 1 shows how the optimisation task is carried out: here an individual encodes the time profile of the ligand concentration outside the cell. While the Fitness Function (FF) is used to estimate the quality of the single individual, mutation and crossover operators boost the search space exploration of the GA. This approach should help the algorithm returning the best solution (individual) to the input optimisation problem by optimising, generation after generation, the fitness value of the invididuals in the population.
Microfluidic device design
Implementing complex timevarying signals is quite simple from a computational point of view; however obtaining realisations of signals with such properties is an active area of research in current microfluidics. Developing geometries that satisfy physical conditions for the generation of signals compliant with the specifications imposed by the theoretical results is not a trivial task. Several alternative solutions have been proposed for signal modulation in microfluidic channels [69–74] being [42, 75, 76] the most recent and advanced contribution in this fields; they are based on diverse physical principles like boundary diffusion controlled by relative velocity (like in H filters [74]), by exciting cells with diverse laminar flows that affect different parts of the cell etc.
In particular two main areas of research arose in this context: on the one hand interesting phenomena in fluid dynamics have been investigated [77, 78] in order to address the problem of cell stimulation via contrained signals [79, 80]; on the other hand the area of Digital Microfluidics has found one of its most active fields of research. In order to implement the signals obtained from the previously described input optimisation task we propose a polydimethilsiloxane (PDMS) based platform for cell stimulation which exploits the signal modulation at the microliter level. This platform has been designed to implement spectral properties of the signals that have been characterised during the theoretical study of the system under investigation: in this way it will became part of the optimal experimental design for cell stimulation.
Declarations
Acknowledgements
This work was partially supported by progetto finalizzato ACC, 'Rete Nazionale di Bioinformatica Oncologica' 2007.
This article has been published as part of BMC Bioinformatics Volume 10 Supplement 12, 2009: Bioinformatics Methods for Biomedical Complex System Applications. The full contents of the supplement are available online at http://www.biomedcentral.com/14712105/10?issue=S12.
Authors’ Affiliations
References
 Hornberg JJ, Bruggeman FJ, Westerhoff HV, Lankelma J: Cancer: A Systems Biology disease. Biosystems 2006, 83: 81–90. 10.1016/j.biosystems.2005.05.014View ArticlePubMedGoogle Scholar
 Burden S, Yarden Y: Neuregulins and their receptors: a versatile signaling module in organogenesis and oncogenesic. Neuron 1997, 18: 847–855. 10.1016/S08966273(00)803244View ArticlePubMedGoogle Scholar
 Yarden Y, Sliwkowski MX: Untangling the ErbB signalling network. Nature Reviews. Molecular Cell Biology 2001, 2: 127–37. 10.1038/35052073View ArticlePubMedGoogle Scholar
 Eastman A, Perez RP: New targets and challenges in the molecular therapeutics of cancer. British Journal of Clinical Pharmacology 2006, 62: 5–14. 10.1111/j.13652125.2006.02720.xPubMed CentralView ArticlePubMedGoogle Scholar
 Sebastian S, Settleman J, Reshkin SJ, Azzariti A, Bellizzi A, Paradiso A: The complexity of targeting EGFR signalling in cancer: from expression to turnover. Biochimica Et Biophysica Acta 2006, 1766: 120–39.PubMedGoogle Scholar
 Mendelsohn J, Baselga J: The EGF receptor family as targets for cancer therapy. Oncogene 2000, 19: 6550–65. 10.1038/sj.onc.1204082View ArticlePubMedGoogle Scholar
 Kholodenko BN, Demin OV, Moehren G, Hoek JB: Quantification of Short Term Signaling by the Epidermal Growth Factor Receptor. J Biol Chem 1999, 274: 30169–30181. 10.1074/jbc.274.42.30169View ArticlePubMedGoogle Scholar
 Oda K, Matsuoka Y, Funahashi A, Kitano H: A comprehensive pathway map of epidermal growth factor receptor signaling. Mol Syst Biol 2005., 1:Google Scholar
 Wiley HS, Shvartsman SY, Lauffenburger DA: Computational modeling of the EGFreceptor system: a paradigm for systems biology. Trends in Cell Biology 2003, 13: 43–50. 10.1016/S09628924(02)000090View ArticlePubMedGoogle Scholar
 Schoeberl B, EichlerJonsson C, Gilles ED, Müller G: Computational modeling of the dynamics of the MAP kinase cascade activated by surface and internalized EGF receptors. 2002, 20: 370–5.Google Scholar
 Hornberg JJ, Binder B, Bruggeman FJ, Schoeberl B, Heinrich R, Westerhoff HV: Control of MAPK signalling: from complexity to what really matters. Oncogene 2005, 24: 5533–42. 10.1038/sj.onc.1208817View ArticlePubMedGoogle Scholar
 Breitling R, Hoeller D: Current challenges in quantitative modeling of epidermal growth factor signaling. FEBS Letters 2005, 579: 6289–94. 10.1016/j.febslet.2005.10.034View ArticlePubMedGoogle Scholar
 Zhang Y, WolfYadlin A, Ross PL, Pappin DJ, Rush J, Lauffenburger DA, White FM: Timeresolved mass spectrometry of tyrosine phosphorylation sites in the epidermal growth factor receptor signaling network reveals dynamic modules. Molecular & Cellular Proteomics: MCP 2005, 4: 1240–50. 10.1074/mcp.M500089MCP200View ArticleGoogle Scholar
 Olsen JV, Blagoev B, Gnad F, Macek B, Kumar C, Mortensen P, Mann M: Global, in vivo, and sitespecific phosphorylation dynamics in signaling networks. Cell 2006, 127: 635–48. 10.1016/j.cell.2006.09.026View ArticlePubMedGoogle Scholar
 Blagoev B, Kratchmarova I, Ong SE, Nielsen M, Foster LJ, Mann M: A proteomics strategy to elucidate functional proteinprotein interactions applied to EGF signaling. Nature Biotechnology 2003, 21: 315–8. 10.1038/nbt790View ArticlePubMedGoogle Scholar
 Hengl S, Kreutz C, Timmer J, Maiwald T: Databased identifiability analysis of nonlinear dynamical models. Bioinformatics 2007, 23: 2612–2618. 10.1093/bioinformatics/btm382View ArticlePubMedGoogle Scholar
 Casey FP, Baird D, Feng Q, Gutenkunst RN, Waterfall JJ, Myers CR, Brown KS, Cerione RA, Sethna JP: Optimal experimental design in an EGFR signaling and downregulation model. qbio/0610024 2006. [http://arxiv.org/abs/qbio/0610024]Google Scholar
 Faller D, Klingmuller U, Timmer J: Simulation Methods for Optimal Experimental Design in Systems Biology. SIMULATION 2003, 79: 717–725. 10.1177/0037549703040937View ArticleGoogle Scholar
 Vance W, Arkin A, Ross J: Determination of causal connectivities of species in reaction networks. Proc Natl Acad Sci U S A 2002, 99: 5816–5821. 10.1073/pnas.022049699PubMed CentralView ArticlePubMedGoogle Scholar
 Samoilov M, Arkin A, Ross J: On the deduction of chemical reaction pathways from measurements of time series of concentrations. Chaos (Woodbury, N.Y.) 2001, 11: 108–114. 10.1063/1.1336499View ArticleGoogle Scholar
 Vlad MO, Arkin A, Ross J: Response experiments for nonlinear systems with application to reaction kinetics and genetics. Proc Natl Acad Sci U S A 2004, 101: 7223–7228. 10.1073/pnas.0402049101PubMed CentralView ArticlePubMedGoogle Scholar
 Flaherty P, Jordan MI, Arkin AP: Robust design of biological experiments. Proceedings of the Neural Information Processing Symposium 2005., 2005:Google Scholar
 Mettetal JT, Muzzey D, GómezUribe C, van Oudenaarden A: The frequency dependence of osmoadaptation in Saccharomyces cerevisiae. Science (New York, N.Y.) 2008, 319: 482–4.View ArticleGoogle Scholar
 Steuer R, Kurths J, Fiehn O, Weckwerth W: Observing and interpreting correlations in metabolomic networks. Bioinformatics 2003, 19: 1019–1026. 10.1093/bioinformatics/btg120View ArticlePubMedGoogle Scholar
 Yue H, Brown M, Knowles J, Wang H, Broomhead DS, Kell DB: Insights into the behaviour of systems biology models from dynamic sensitivity and identifiability analysis: a case study of an NF[small kappa]B signalling pathway. Molecular BioSystems 2006, 2: 640–649. 10.1039/b609442bView ArticlePubMedGoogle Scholar
 Quaiser T, Marquardt W, Monnigmann M: Local identifiability analysis of large signalling pathway models. Proc. of FOSBE 2007 2007, 465–470.Google Scholar
 Anguelova M, Wennberg B: Identifiability of the Timelag Parameter in Delay Systems with Applications to Systems Biology. Proc. of FOSBE 2007 2007, 533–538.Google Scholar
 Geffen Dara, S M, A F, Findeisen Rolf, Guay M: The Question of Parameter Identifiability for Biochemical Reaction Networks Considering the NFkappaB Signal Transduction Pathway. Proc. of FOSBE 2007 2007, 509–514.Google Scholar
 Cobelli C, RomaninJacur G: Structural identifiability of strongly connected biological compartmental systems. Med Biol Eng 1975, 13: 831–838. 10.1007/BF02478086View ArticlePubMedGoogle Scholar
 Cobelli C, Lepschy A, RomaninJacur G:Structural identifiability of biological compartmental systems. Digital computer implementation of a testing procedure. 1976, 88–102. [http://dx.doi.org/10.1007/3–540–07622–0_463]Google Scholar
 Bellu G, Saccomani MP, Audoly S, D'Angiò L: DAISY: a new software tool to test global identifiability of biological and physiological systems. Comput Methods Programs Biomed 2007, 88: 52–61. 10.1016/j.cmpb.2007.07.002PubMed CentralView ArticlePubMedGoogle Scholar
 Pronzato L: Optimal experimental design and some related control problems. 0802.4381 2008. [Automatica/Automatica J IFAC; Automatika; Automatica IFAC J 44 (2008) 303–325]. [http://arxiv.org/abs/0802.4381] [Automatica/Automatica J IFAC; Automatika; Automatica IFAC J 44 (2008) 303–325].Google Scholar
 Ljung L: System Identification: Theory for the User. 2nd edition. Prentice Hall PTR; 1999.Google Scholar
 Phair RD, Misteli T: Kinetic modelling approaches to in vivo imaging. Nature Reviews. Molecular Cell Biology 2001, 2: 898–907. [PMID: 11733769]. [PMID: 11733769]. 10.1038/35103000View ArticlePubMedGoogle Scholar
 Cho KH, Shin SY, Kolch W, Wolkenhauer O: Experimental Design in Systems Biology, Based on Parameter Sensitivity Analysis Using a Monte Carlo Method: A Case Study for the TNFalphaMediated NFkappa B Signal Transduction Pathway. SIMULATION 2003, 79: 726–739. 10.1177/0037549703040943View ArticleGoogle Scholar
 Lipschultz CA, Li Y, SmithGill S: Experimental design for analysis of complex kinetics using surface plasmon resonance. Methods (San Diego, Calif.) 2000, 20: 310–8.View ArticleGoogle Scholar
 Sontag E: Molecular Systems Biology and Control: A QualitativeQuantitative Approach. Decision and Control, 2005 and 2005 European Control Conference. CDCECC '05. 44th IEEE Conference on 2005 2314–2319.
 Maiwald T, Timmer J: Dynamical Modeling and MultiExperiment Fitting with PottersWheel. Bioinformatics 2008, btn350. [http://bioinformatics.oxfordjournals.org/cgi/content/abstract/btn350v1]Google Scholar
 BalsaCanto E, Alonso A, Banga J: Computational procedures for optimal experimental design in biological systems. IET Systems Biology 2008, 2(4):163–172. 10.1049/ietsyb:20070069View ArticlePubMedGoogle Scholar
 Rylander BI: Computational complexity and the genetic algorithm. PhD thesis. University of Idaho; 2001. [http://portal.acm.org/citation.cfm?id=933869]Google Scholar
 Hollander M, Wolfe DA: Nonparametric Statistical Methods. 2nd edition. WileyInterscience; 1999.Google Scholar
 Andrew N, Craig D, Urbanski JP, Gunawardena J, Thorsen T: Microfluidic temporal cell stimulation. μTAS 08 2008.Google Scholar
 Kholodenko BN, Kiyatkin A, Bruggeman FJ, Sontag E, Westerhoff HV, Hoek JB: Untangling the wires: a strategy to trace functional interactions in signaling and gene networks. Proceedings of the National Academy of Sciences of the United States of America 2002, 99(20):12841–6. 10.1073/pnas.192442699PubMed CentralView ArticlePubMedGoogle Scholar
 Sontag E, Kiyatkin A, Kholodenko BN: Inferring dynamic architecture of cellular networks using time series of gene expression, protein and metabolite data. Bioinformatics 2004, 20(12):1877–1886. 10.1093/bioinformatics/bth173View ArticlePubMedGoogle Scholar
 Thies W, Urbanski J, Thorsen T, Amarasinghe S: Abstraction layers for scalable microfluidic biocomputing. Natural Computing 2006. [http://dx.doi.org/10.1007/s11047–006–9032–6]Google Scholar
 BioModels Database[http://www.ebi.ac.uk/biomodelsmain/staticpages.do?page=home]
 BioModels Database – Kholodenko1999EGFRsignaling[http://www.ebi.ac.uk/biomodelsmain/publmodel.do?mid=BIOMD0000000048]
 Hoops S, Sahle S, Gauges R, Lee C, Pahle J, Simus N, Singhal M, Xu L, Mendes P, Kummer U: COPASIa COmplex PAthway SImulator. Bioinformatics 2006, 22: 3067–3074. 10.1093/bioinformatics/btl485View ArticlePubMedGoogle Scholar
 Gunawardena J: Signals and Systems: Towards a Systems Biology of Signal Transduction. Proceedings of the IEEE 2008, 96: 1386–1397. 10.1109/JPROC.2008.925413View ArticleGoogle Scholar
 Hu S: Optimal time points sampling in pathway modelling. Engineering in Medicine and Biology Society, 2004. IEMBS '04. 26th Annual International Conference of the IEEE 2004, 1: 671–674.Google Scholar
 Steinke F, Seeger M, Tsuda K: Experimental design for efficient identification of gene regulatory networks using sparse Bayesian models. BMC Systems Biology 2007, 1: 51. 10.1186/17520509151PubMed CentralView ArticlePubMedGoogle Scholar
 van Riel NA: Dynamic modelling and analysis of biochemical networks: mechanismbased models and modelbased experiments. Brief Bioinform 2006, 7: 364–374. 10.1093/bib/bbl040View ArticlePubMedGoogle Scholar
 Asprey S, Macchietto S: Designing robust optimal dynamic experiments 2002.[http://www.ingentaconnect.com/content/els/09591524/2002/00000012/00000004/art00020]
 Apgar JF, Toettcher JE, Endy D, White FM, Tidor B: Stimulus Design for Model Selection and Validation in Cell Signaling. PLoS Computational Biology 2008, 4: e30. 10.1371/journal.pcbi.0040030PubMed CentralView ArticlePubMedGoogle Scholar
 Bernaerts K, Impe JV: Optimal dynamic experiment design for estimation of microbial growth kinetics at suboptimal temperatures: Modes of implementation. Simulation Modelling Practice and Theory 2005, 13: 129–138. 10.1016/j.simpat.2004.10.005View ArticleGoogle Scholar
 Kremling A, Fischer S, Gadkar K, Doyle FJ, Sauter T, Bullinger E, Allgower F, Gilles ED: A Benchmark for Methods in Reverse Engineering and Model Discrimination: Problem Formulation and Solutions. Genome Res 2004, 14: 1773–1785. 10.1101/gr.1226004PubMed CentralView ArticlePubMedGoogle Scholar
 Brik Ternbach M, Bollman C, Wandrey C, Takors R: Application of model discriminating experimental design for modeling and development of a fermentative fedbatch Lvaline production process. Biotechnol Bioeng 2005, 91: 356–368. 10.1002/bit.20504View ArticlePubMedGoogle Scholar
 Franceschini G, Sanro Macchietto: Modelbased design of experiments for parameter precision: State of the art. Chemical Engineering Science 2007.Google Scholar
 Chen B, Asprey S: On the Design of Optimally Informative Dynamic Experiments for Model Discrimination in Multiresponse Nonlinear Situations. Industrial & Engineering Chemistry Research 2003, 42: 1379–1390. 10.1021/ie0203025View ArticleGoogle Scholar
 Cooney M, McDonald K: Optimal dynamic experiments for bioreactor model discrimination. Applied Microbiol Biotechnol 1995, 43: 826–837. 10.1007/BF02431915View ArticleGoogle Scholar
 Atkinson AC, Donev AN: Optimum experimental designs. Optimum Experimental Designs
 Casey F: Prediction and Optimal Experimental Design in Systems Biology. In PhD in Physics. Cornell University, LASSP, Physics Department, Clark Hall, Cornell University, Ithaca, NY 14853–2501; 2006.Google Scholar
 UCSB Biosens – BioSPICE Dashboard[http://www.chemengr.ucsb.edu/~ceweb/faculty/doyle/biosens/BioSens.htm]
 UCSB Biosens – BioSPICE Dashboard[http://cbbl.imim.es:8080/ByoDyn]
 Gutenkunst RN, Waterfall JJ, Casey FP, Brown KS, Myers CR, Sethna JP: Universally Sloppy Parameter Sensitivities in Systems Biology Models. PLoS Comput Biol 2007, 3(10):1871–1878. 10.1371/journal.pcbi.0030189View ArticlePubMedGoogle Scholar
 Mehra R: Optimal input signals for parameter estimation in dynamic systemsSurvey and new results. Automatic Control, IEEE Transactions on 1974, 19(6):753–768. 10.1109/TAC.1974.1100701View ArticleGoogle Scholar
 RodriguezFernandez M, Egea J, Banga J: Novel metaheuristic for parameter estimation in nonlinear dynamic biological systems. BMC Bioinformatics 2006, 7: 483. 10.1186/147121057483PubMed CentralView ArticlePubMedGoogle Scholar
 Goodwin G: Identification: Experiment Design. Systems and Control Encyclopedia 1987, 4: 2257–64.Google Scholar
 Weibel DB, Whitesides GM: Applications of microfluidics in chemical biology. Current Opinion in Chemical Biology 2006, 10: 584–591. 10.1016/j.cbpa.2006.10.016View ArticlePubMedGoogle Scholar
 Walker GM, Zeringue HC, Beebe DJ: Microenvironment design considerations for cellular scale studies. Lab on a Chip 2004, 4: 91–97. 10.1039/b311214dView ArticlePubMedGoogle Scholar
 Takayama S, Ostuni E, LeDuc P, Naruse K, Ingber DE, Whitesides GM: Selective Chemical Treatment of Cellular Microdomains Using Multiple Laminar Streams. Chemistry & Biology 2003, 10: 123–130. 10.1016/S10745521(03)00019XView ArticleGoogle Scholar
 Takayama S, Ostuni E, LeDuc P, Naruse K, Ingber DE, Whitesides GM: Subcellular positioning of small molecules. Nature 2001, 411: 1016. 10.1038/35082637View ArticlePubMedGoogle Scholar
 Breslauer DN, Lee PJ, Lee LP: Microfluidicsbased systems biology. Molecular BioSystems 2006, 2: 97–112. 10.1039/b515632gView ArticlePubMedGoogle Scholar
 Squires TM, Quake SR: Microfluidics: Fluid physics at the nanoliter scale. Reviews of Modern Physics 2005, 77: 977–50. 10.1103/RevModPhys.77.977View ArticleGoogle Scholar
 Ye L, Zhang M, Alexopoulosa L, Sorger P, Jensen K: Microfluidic devices for studying the response of adherent cells under short time stimuli treatment. μTAS 07 2007.Google Scholar
 King KR, Wang S, Jayaraman A, Yarmush ML, Toner M: Microfluidic flowencoded switching for parallel control of dynamic cellular microenvironments. Lab on a Chip 2008, 8: 107–116. 10.1039/b716962kView ArticlePubMedGoogle Scholar
 Urbanski JP: Application of Microfluidic Emulsion Technology to Biochemistry, Drug Delivery and LabonaChip Programmability. 2005.Google Scholar
 Thorsen T: Microfluidic Technologies for HighThroughput Screening Applications. 2003.Google Scholar
 Thomson TM: TMT Thesis Project – OpenWetWare.2005. [http://openwetware.org/wiki/TMT_Thesis_Project]Google Scholar
 Thomson TM: Stimulator Project – OpenWetWare.2005. [http://openwetware.org/wiki/Stimulator]Google 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.