The Impact of Time Delays on the Robustness of Biological Oscillators and the Effect of Bifurcations on the Inverse Problem
© The Author(s) 2009
Received: 30 May 2008
Accepted: 14 August 2008
Published: 2 November 2008
Differential equation models for biological oscillators are often not robust with respect to parameter variations. They are based on chemical reaction kinetics, and solutions typically converge to a fixed point. This behavior is in contrast to real biological oscillators, which work reliably under varying conditions. Moreover, it complicates network inference from time series data. This paper investigates differential equation models for biological oscillators from two perspectives. First, we investigate the effect of time delays on the robustness of these oscillator models. In particular, we provide sufficient conditions for a time delay to cause oscillations by destabilizing a fixed point in two-dimensional systems. Moreover, we show that the inclusion of a time delay also stabilizes oscillating behavior in this way in larger networks. The second part focuses on the inverse problem of estimating model parameters from time series data. Bifurcations are related to nonsmoothness and multiple local minima of the objective function.
The investigation of regulation mechanisms underlying various properties of cellular networks has gained much attention in recent years. Especially interesting in this setting is the relation between the topology of a regulatory network, often referred to as wiring diagram or interaction graph, and the ability of the system to exhibit certain kinds of dynamic behaviors. It is well known that feedback control mechanisms are essential for phenomena such as hysteresis, bistability, multistationarity, and periodic behavior (see, e.g., [1â€“, 4] or  for a more recent review).
While these feedback mechanisms are necessary to capture such phenomena, their existence is usually by no means sufficient. A strong nonlinearity in the Goodwin oscillator model, for example, is a very restrictive requirement for oscillations [6, 7]. Cooperative interaction is also needed to capture switch-like behavior between two or more stable steady states [6, 8â€“, 10]. Thus, the qualitative behavior of the system depends considerably on the exact parameter values . Periodic behavior, for example, can often only be observed for a small fraction in the parameter space, which is bounded by bifurcation manifolds .
This is in contrast to real biological systems, which exhibit their function reliably under varying external conditions and internal noise [13â€“, 15], raising the question how this robustness is achieved. The design principles of biological networks are assumed to be a result of a long evolutionary process [16, 17], during which the principles are optimized for a reliable functioning. Signaling networks, for example, have to be sensitive to signals and robust against random perturbations and internal fluctuations at the same time [18â€“, 20]. Many cellular oscillators such as the circadian clock have to maintain a constant period and amplitude under a wide range of different external conditions . The regulatory network underlying the cell cycle has to be robust against perturbations, since dysfunctions may lead to programmed cell death or to phenotypes that are not able to survive for a long time, if at all [22, 23].
All these biological examples investigate a property of organisms which can be described by functional robustness . However, the exact definition of robustness varies in all these publications, which indicates that a formalization of the concept of robustness has not yet been established [15, 24]. Further, this goes along with the question about which mechanisms are potentially related to such a robustness.
In this paper, we focus on the robustness of biological oscillator models with respect to varying model parameters. Time-scale differences, time delays, and, related to that, feedback loops comprising a large number of interactions, have already been shown to maintain periodic behavior in chemical reaction systems (see [1, 25] and references therein). Scheper et al. , for example, demonstrated the importance of nonlinear regulation and time delays on a model of the circadian oscillator. Chen and Aihara  investigated the effect of large time-scale differences and time delays on a two-component oscillator model. Generalizations of their results can be found in . A stabilization of oscillations via time delays among others has also been reported in [21, 28, 29]. While many of the earlier studies refer to two-component systems, interesting recent studies indicate the impact of multiple interlocked feedback loops for the robustness of periodic behavior [30â€“, 35].
This work focuses on oscillations induced by including a time delay into the differential equation model. This inclusion can destabilize a stable fixed point by a Hopf bifurcation. An ordinary differential equation (ODE) model describes the cell as a homogeneous chemical reaction system, assuming that the time between cause and effect of a regulatory process can be neglected. This is of course a simplification, since time delays play a role in many regulation processes. Examples are the transport of mRNA from the nucleus to the cytoplasm, diffusion processes, especially in eukaryotic cells, or the time between binding of a transcription factor to the DNA and the corresponding change in concentration of the regulated gene product. The inclusion of such time delays into the ODE models can change the dynamic behavior of the system qualitatively. Furthermore, the period of oscillations has been shown to be crucially regulated by such a delay .
The model class considered here is characterized by monotonicity and boundedness constraints, which will be explained in detail in Section 2.1. This class is similar to systems investigated by Kaufman et al.  and Pigolotti et al. . Since the proofs rely on very weak assumptions about the differential equation system, they apply to many two-component oscillator models which have already extensively been studied (see, e.g., [6, 25, 26]). In this sense, the paper generalizes some of the previous publications.
Section 2.2 shows results for two-dimensional systems. In particular, sufficient conditions for the destabilization of a steady state via a time delay are introduced, which imply the existence of a stable limit cycle.
Higher dimensional feedback systems are studied in Section 2.3. For a single negative feedback, system I shows that the inclusion of a time delay can destabilize a stable fixed point through a Hopf bifurcation, implying oscillating behavior. In turn, an unstable fixed point cannot become stable through a time delay.
Section 3 elucidates the problem of robustness of oscillations from a different point of view, the inference of oscillating models from time series data. We will demonstrate on a two-gene network that bifurcations complicate parameter estimation considerably. They are related to nonsmooth error functions with multiple local optima. The special focus in this study is on the bifurcations relevant for chemical oscillator models. As already pointed out by several authors (see, e.g., [37â€“, 40]), results emphasize that advanced parameter estimation approaches for differential equations are required in this context. Finally, conclusions and ideas for future work are provided in Section 4.
2. Stabilizing Oscillations with Time Delays
2.1. Modeling Biological Oscillators
with a continuously differentiable function . The function is characterized by the monotonicity of each component with respect to . This condition assigns each regulator of either a purely activating or a purely inhibiting function. In the first case, independent of the state of the system, the second case corresponds to for all states . Such systems can be illustrated by directed graphs with sign-labeled edges, often denoted interaction graphs, or, equivalently, by the signed Jacobian matrix with elements according to the edge signs in . Properties of this model class and the role of feedback loops are discussed in [1, 2, 4].
Moreover, we assume solutions of the system to be bounded. This is a biologically plausible assumption, but excludes simple linear models. This boundedness constraint is, for example, fulfilled for all network models which describe degradation of network components as a first-order decay process and assume bounds for the production rate . Solutions of these systems have the tendency to converge to steady states [41, 42]. Thus, more complex behavior such as oscillations is typically caused by destabilizing this steady state via Hopf bifurcations . Such a bifurcation requires the existence of a negative feedback loop in . Hence we focus the analysis onto the investigation of the stability of fixed points , which can be done via investigating the eigenvectors of the Jacobian matrix . Elements of will be denoted by throughout the manuscript, dropping their dependence on the coordinates of . According to Lyapunov's indirect method , a hyperbolic steady-state is stable if all eigenvalues of have negative real parts [44, 45]. This statement also holds for differential equations including time delays . Further, we will refer to conditions that imply periodic behavior when a fixed point is destabilized.
2.2. Stabilizing Oscillations with Time Delays in Two-Dimensional Systems
Equation (5) is a polynomial of degree two for , which has two complex conjugate solutions . For , it is a transcendental equation with a countable infinite number of roots. However, the number of roots in the right-half plane is known to be finite . Here, we will only investigate the course of the two solutions in dependence of .
A stable limit cycle surrounds an unstable fixed point that is given as the intersection of the two nullclines. A fixed point can only be unstable if it is located between the minimum and the maximum of nullcline 1. This corresponds to a positive element at the fixed point . It has been shown that this condition becomes sufficient for sufficiently large time-scale differences [26, 27]. Furthermore, it is not required any more when including time delays, which was exemplarily shown for a specific AIO model in .
Here, we show that a stable fixed point with signed Jacobian matrix of the forms or (6) can indeed always be destabilized through a time delay . This result is an extension of the previous results described in , where we proved that a bifurcation via an increase in the time delay always destabilizes a fixed point. Furthermore, this destabilization is caused by a Hopf bifurcation and therefore creates a stable limit cycle. Thus, becomes as well a sufficient condition for the existence of a stable limit cycle in AIO models, provided that the time delay is sufficiently large. This is stated in the following Theorem 1. The proof is given in Appendix A. We remark here that this theorem can analogously be proven for SDO models.
Theorem 1 (instability of through a time delay).
Assume system (2) to have a stable fixed point for and signed Jacobian matrix as given in (6). An increase in the time delay eventually destabilizes .
Theorem 1 implies that there exists a threshold time delay such that is unstable if .
If system (2) has a stable fixed point for and , there exists a threshold time delay such that is unstable for . The proof is given in Appendix B
According to Theorem 1 and Corollary 1, if an AIO model has a single fixed point that lies between the minimum and the maximum of the first nullcline (Figure 1), this fixed point is unstable for a sufficiently large time delay .
The inclusion of a sufficiently large time delay like in (2) destabilizes via a Hopf bifurcation. This guarantees the existence of a stable limit cycle at least around the bifurcation value . Moreover, in case that is the only fixed point of the system, destabilizing implies global convergence to a stable limit cycle from arbitrary initial conditions. This follows from the PoincarÃ©-Bendixson theorem (PBT), which states that the -limit set of a bounded forward trajectory of a two-dimensional system is either a steady-state or a limit cycle . In other words, a bounded solution of such a system either converges to a fixed point or to a limit cycle.
2.3. Generalizations for Higher Dimensions
The following theorem generalizes results in , where the same statement was proven for two-dimensional systems.
The proof can be found in Appendix C.
Theorem 2 shows that whenever an increase in the time delay causes a Hopf bifurcation in the single loop system, this bifurcation destabilizes a stable fixed point and creates a limit cycle.
However, sufficient conditions for the occurrence of such a bifurcation in negative loop systems similar to that in two-dimensional systems remain to be investigated in this context. Also for the single loop system, the introduction of a positive autoregulation of one of the components seems to be sufficient for the existence of such a threshold value, with the same convergence argument as for two-dimensional systems. Unlike in two-dimensional systems, however, it is not clear for arbitrary network structures whether the destabilization of a fixed point implies the existence of a stable limit cycle not only locally in the neighborhood of a bifurcation.
3. On the Impact of Bifurcations for the Inverse Problem
Optimization problems of this kind are important for all fields in which differential equations are used to describe dynamic behaviors, and model parameters are to be adapted to experimental data. For nonlinear systems, these problems are known to be difficult to solve, since the surface of the objective function has some undesirable properties. Efficient optimization algorithms are required in order to obtain reliable estimates within an acceptable time. Several approaches have been proposed (see, e.g.,  and the subsequent discussions for an overview, or [37, 38] for a method called “multiple shooting” and applications to biological systems). Generally, these optimization problems will become even more important in systemsâ€™ biology in the future.
with parameter vector . The true vector is given by . The system has a globally stable limit cycle for these values.
In the limit , (13) corresponds to the area between the two “curves” and . Values for are numerically calculated by a simple Euler discretization, and initial values are set to . By the way, numerical integration has to be performed in each step of a gradient-based optimization approach and is usually the limiting factor concerning computing time.
Of course, even without noise, (12) and (13) depend on the dataset , in particular, on the initial vector and the sampling time points. Here, we show exemplary examples for fixed initial conditions and simulations over the transient and two oscillation periods. Further, for simplicity reasons, we vary only one single parameter at a time, the control parameter, while the rest is fixed to the true values . Thus, the measurements are obtained via simulations using , and corresponds to simulations using and a different value for the control parameter . In order to overcome the dependence of the error functions (12) and (13) on the sequence of sampling time points, we use a very small time step , which corresponds to 10.000 Euler steps for numerical integration in the simulated range.
Such jumps in the error function are generally related to bifurcations at which stable -limit sets disappear, typically saddle-node or subcritical Hopf bifurcations in our context. As a consequence, gradient-based methods might be much more efficient when the step size is adapted during the gradient descent to optimize .
3.2. Local Suboptimal Minima
Figure 10 shows the bifurcation diagram with control parameter . For low values of , the system has a stable limit cycle around an unstable fixed point. This limit cycle vanishes at a saddle-node bifurcation (SN), where an unstable and a globally stable fixed point emerges. While the dependence of the -coordinate of this fixed point is only marginal (Figure 11(a)), the coordinates of increase with increasing (Figure 11(b)), which leads to a local suboptimal minimum in the error functions, here at a value (Figure 12). Moreover, it can be seen that the true value has a relatively small basin of attraction, which is bounded by the saddle-node bifurcation (SN). On the contrary, the basin of attraction for the local minimum at , which corresponds to the converging time series, is much larger.
Hence, starting a local search method with an arbitrary initial parameter vector leads in most cases to suboptimal minima which correspond to systems that converge to a stable fixed point. These local minima render global search methods such as simulated annealing or genetic algorithms necessary, which usually require long running times . Thus, efficient algorithms are needed in this context.
3.3. Ruggedness Near the True Parameter Value
Figure 13 shows the bifurcation diagram with control parameter . The system has a stable limit cycle bounded by two supercritical Hopf bifurcations (HBs). Within this oscillating region, period and amplitude vary considerably with the control parameter, as indicated in the simulations in Figure 14. This dependency causes multiple local minima in the error functions (Figure 15). Thus, even if the optimization process is already started within the oscillating region in the parameter space, a simple gradient search might fail to find the true parameter value but get stuck in one of the local minima.
This emphasizes again the necessity of efficient optimization algorithms for parameter estimation of differential equation models in general.
This paper investigated the robustness of sustained oscillations in regulatory systems with respect to varying model parameters. Differential equations based on chemical reaction kinetics, which are often used for this purpose, are not always robust, and oscillations only occur in a small region of the parameter space bounded by bifurcation manifolds.
In the first part of the paper, we focused on the inclusion of time delays into the differential equations. Time delays take some time between the cause and the effect of a regulation into account, and they are known to stabilize oscillations by enlarging the region in the parameter space which correspond to periodic solutions. Since the typical behavior of the class of systems considered here is convergence to a fixed point, oscillations are usually induced via destabilizing a fixed point through a Hopf bifurcation. We investigated the stability of a fixed point in dependence of the time delay. We provided sufficient conditions for a time delay to induce oscillations in two-dimensional systems, in particular, activator-inhibitor oscillator and substrate-depletion oscillator models, which are the typical oscillator models in two dimensions. These conditions are graphically defined in terms of the qualitative course of the nullclines, which are usually easily accessible. Specifically, if the system has a single fixed point located between the minimum and the maximum of one of the nullclines, it can always be destabilized by a sufficiently large time delay, which implies sustained oscillations. Results are based on rather general assumptions about the underlying differential equation system, which hold for many related oscillator models.
Moreover, for single-loop systems with an arbitrary number of components we showed that a Hopf bifurcation that is caused by increasing the time delay always destabilizes a stable fixed point. The real parts of the eigenvalues of the Jacobian matrix at the fixed point change signs from negative to positive. Thus, a stable fixed point can loose stability by increasing the time delay, which leads to the existence of a stable limit cycle, but an unstable fixed point cannot become stable.
Here, the analysis of the system was done by linearizing the system about a fixed point and investigating the stability of this fixed point via the spectrum of eigenvalues. This facilitates the analysis of the long-term behavior considerably. We referred to the conditions necessary for such a destabilization of a fixed point to imply sustained oscillations. The PoincarÃ©-Bendixson theorem is extremely useful in this context for two-dimensional systems. Similar theorems exist for higher-dimensional systems with special interaction graphs. The single-loop system considered here belongs to these systems. However, for a further generalization of these results to higher dimensional systems, the following questions remain to be investigated in the future. First, can the class of systems that have a unique fixed point be further characterized? It is already known that networks lacking a positive feedback loop have at most one fixed point. Second, how can this be further generalized to networks that also contain positive feedback circuits? Networks with only positive loops cannot have stable limit cycles. Consequently, oscillations can only occur in networks that have at least one negative loop. Contrary to negative feedback control, positive loops can lead to multiple fixed points. Hence for such “mixed-circuit networks,” it is not sufficient any more to show the existence of a fixed point. It also has to be investigated whether it is unique. However, necessary conditions for multiple fixed points in positive loop systems are also known to be very restrictive, and many of these models seem to have a unique fixed point, too. Third, in which cases does a destabilization of a fixed point lead to oscillating behavior? And forth, what are sufficient conditions for the existence of a threshold time delay ?
The second part of this work investigated the influence of bifurcations on the inverse problem to estimate model parameters from time series data. Such bifurcations are generally related to nonsmoothness and multiple local minima of the objective function to be optimized in this setting. Although these phenomena are generally not new, the focus was on the special properties occurring in the class of oscillating models considered here. Global search methods are required to find the real optimum. Together with the numerical integration, these methods are usually extremely time-consuming, even for small systems with only a few parameters.
In a realistic setting, the problem is even worse. First, the optimization problem is of course multidimensional. All values of model parameters have in principle to be found at a time, which renders a comprehensive search and an investigation of the whole objective function difficult. Second, the data is usually noisy and sparse, leading to ill-posed optimization problems. Noisy datasets also mean that the measured initial condition might not always be the best choice for . Generally, should be included into the objective function as an additional variable that has to be optimized as well, which increases the dimension of the inverse problem even further. Moreover, the dataset might also contain missing values or unobserved variables. This raises additional problems, and estimating parameters by minimizing the residual error might fail in this context anyway. In this setting, stochastic approaches might be more convenient, since they take the noise in the dataset into account. Bayesian learning approaches, for example, allow for an appropriate regularization via prior distributions over model parameters. Concluding, the development of efficient approaches for parameter estimation in differential equation models remains a challenging research field in the future.
- Cinquin O, Demongeot J: Roles of positive and negative feedback in biological systems. Comptes Rendus Biologies 2002, 325(11):1085-1095. 10.1016/S1631-0691(02)01533-0View ArticleGoogle Scholar
- GouzÃ© J-L: Positive and negative circuits in dynamical systems. Journal of Biological System 1998, 6(21):11-15.View ArticleGoogle Scholar
- Snoussi EH: Necessary conditions for multistationarity and stable periodicity. Journal of Biological System 1998, 6(1):3-9. 10.1142/S0218339098000042View ArticleMATHGoogle Scholar
- Thomas R, D'Ari R: Biological Feedback. CRC Press, Boca Raton, Fla, USA; 1990.MATHGoogle Scholar
- Thieffry D: Dynamical roles of biological regulatory circuits. Briefings in Bioinformatics 2007, 8(4):220-225. 10.1093/bib/bbm028View ArticleGoogle Scholar
- Fall CP, Marland ES, Wagner JM, Tyson JJ (Eds): Computational Cell Biology, Interdisciplinary Applied Mathematics. Volume 20. Springer, New York, NY, USA; 2005.Google Scholar
- Goodwin BC: Oscillatory behavior in enzymatic control processes. Advances in Enzyme Regulation 1965, 3: 425-438.View ArticleGoogle Scholar
- Angeli D, Ferrell JE Jr., Sontag ED: Detection of multistability, bifurcations, and hysteresis in a large class of biological positive-feedback systems. Proceedings of the National Academy of Sciences of the United States of America 2004, 101(7):1822-1827. 10.1073/pnas.0308265100View ArticleGoogle Scholar
- Craciun G, Tang Y, Feinberg M: Understanding bistability in complex enzyme-driven reaction networks. Proceedings of the National Academy of Sciences of the United States of America 2006, 103(23):8697-8702. 10.1073/pnas.0602767103View ArticleMATHGoogle Scholar
- Kaufman M, SoulÃ© C, Thomas R: A new necessary condition on interaction graphs for multistationarity. Journal of Theoretical Biology 2007, 248(4):675-685. 10.1016/j.jtbi.2007.06.016View ArticleMathSciNetGoogle Scholar
- EiÃŸing T, Waldherr S, AllgÃ¶wer F, Scheurich P, Bullinger E: Steady state and (bi-) stability evaluation of simple protease signalling networks. Biosystems 2007, 90(3):591-601. 10.1016/j.biosystems.2007.01.003View ArticleGoogle Scholar
- Goldbeter A: Computational approaches to cellular rhythms. Nature 2002, 420(6912):238-245. 10.1038/nature01259View ArticleGoogle Scholar
- Barkai N, Leibler S: Circadian clocks limited by noise. Nature 2000, 403(6767):267-268.Google Scholar
- Kitano H: Systems biology: a brief overview. Science 2002, 295(5560):1662-1664. 10.1126/science.1069492View ArticleGoogle Scholar
- Stelling J, Sauer U, Szallasi Z, Doyle FJ, Doyle J: Robustness of cellular functions. Cell 2004, 118(6):675-685. 10.1016/j.cell.2004.09.008View ArticleGoogle Scholar
- Kollmann M, LÃ¸vdok L, BartholomÃ© K, Timmer J, Sourjik V: Design principles of a bacterial signalling network. Nature 2005, 438(7067):504-507. 10.1038/nature04228View ArticleGoogle Scholar
- Ma W, Lai L, Ouyang Q, Tang C: Robustness and modular design of the Drosophila segment polarity network. Molecular Systems Biology 2006, 2, article 70: 1-9.Google Scholar
- Hornung G, Barkai N: Noise propagation and signaling sensitivity in biological networks: a role for positive feedback. PLoS Computational Biology 2008, 4(1):e8. 10.1371/journal.pcbi.0040008View ArticleMathSciNetGoogle Scholar
- Komarova NL, Zou X, Nie Q, Bardwell L: A theoretical framework for specificity in cell signaling. Molecular Systems Biology 2005, 1, article 2005.0023: 1-5.View ArticleGoogle Scholar
- Kwon Y-K, Cho K-H: Quantitative analysis of robustness and fragility in biological networks based on feedback dynamics. Bioinformatics 2008, 24(7):987-994. 10.1093/bioinformatics/btn060View ArticleGoogle Scholar
- Wang R, Zhou T, Jing Z, Chen L: Modelling periodic oscillation of biological systems with multiple timescale networks. IEE Proceedings Systems Biology 2004, 1(1):71-84.View ArticleGoogle Scholar
- BÃ¤hler J: Cell-cycle control of gene expression in budding and fission yeast. Annual Review of Genetics 2005, 39: 69-94. 10.1146/annurev.genet.39.110304.095808View ArticleGoogle Scholar
- Chen KC, Calzone L, Csikasz-Nagy A, Cross FR, Novak B, Tyson JJ: Integrative analysis of cell cycle control in budding yeast. Molecular Biology of the Cell 2004, 15(8):3841-3862. 10.1091/mbc.E03-11-0794View ArticleGoogle Scholar
- Kitano H: Towards a theory of biological robustness. Molecular Systems Biology 2007., 3, article 137:Google Scholar
- Scheper T, Klinkenberg D, Pennartz C, van Pelt J: A mathematical model for the intracellular circadian rhythm generator. The Journal of Neuroscience 1999, 19(1):40-47.Google Scholar
- Chen L, Aihara K: A model of periodic oscillation for genetic regulatory systems. IEEE Transactions on Circuits and Systems I 2002, 49(10):1429-1436. 10.1109/TCSI.2002.803354View ArticleMathSciNetGoogle Scholar
- Radde N: The effect of time scale differences and time-delays on the structural stability of oscillations in a two-gene network. Advances in Complex Systems 2008, 11(3):471-483. 10.1142/S0219525908001751View ArticleMathSciNetMATHGoogle Scholar
- BuriÄ‡ N, TodoroviÄ‡ D: Dynamics of delay-differential equations modelling immunology of tumor growth. Chaos, Solitons & Fractals 2002, 13(4):645-655. 10.1016/S0960-0779(00)00275-7View ArticleGoogle Scholar
- Schmitz S, Loeffler M, Jones JB, Lange RD, Wichmann HE: Synchrony of bone marrow proliferation and maturation as the origin of cyclic haemopoiesis. Cell and Tissue Kinetics 1990, 23(5):425-442.Google Scholar
- Clodong S, DÃ¼hring U, Kronk L, et al.: Functioning and robustness of a bacterial circadian clock. Molecular Systems Biology 2007, 3, article 90: 1-9.Google Scholar
- Kuczenski RS, Hong KC, GarcÃa-Ojalvo J, Lee KH: PERIOD-TIMELESS interval timer may require an additional feedback loop. PLoS Computational Biology 2007, 3(8):e154. 10.1371/journal.pcbi.0030154View ArticleMathSciNetGoogle Scholar
- Locke JC, Southern MM, Kozma-BognÃ¡r L, et al.: Extension of a genetic network model by iterative experimentation and mathematical analysis. Molecular Systems Biology 2005, 1, article 2005.0013: 1-9.View ArticleGoogle Scholar
- Romond P-C, Rustici M, Gonze D, Goldbeter A: Alternating oscillations and chaos in a model of two coupled biochemical oscillators driving successive phases of the cell cycle. Annals of the New York Academy of Sciences 1999, 879: 180-193. 10.1111/j.1749-6632.1999.tb10419.xView ArticleGoogle Scholar
- Venkatesh KV, Bhartiya S, Ruhela A: Multiple feedback loops are key to a robust dynamic performance of tryptophan regulation in Escherichia coli . FEBS Letters 2004, 563(1â€“3):234-240.View ArticleGoogle Scholar
- Wagner A: Circuit topology and the evolution of robustness in two-gene circadian oscillators. Proceedings of the National Academy of Sciences of the United States of America 2005, 102(33):11775-11780. 10.1073/pnas.0501094102View ArticleGoogle Scholar
- Pigolotti S, Krishna S, Jensen MH: Oscillation patterns in negative feedback loops. Proceedings of the National Academy of Sciences of the United States of America 2007, 104(16):6533-6537. 10.1073/pnas.0610759104View ArticleMathSciNetMATHGoogle Scholar
- Balsa-Canto E, Peifer M, Banga JR, Timmer J, Fleck C: Hybrid optimization method with general switching strategy for parameter estimation. BMC Systems Biology 2008, 2, article 26: 1-9.Google Scholar
- Peifer M, Timmer J: Parameter estimation in ordinary differential equations for biochemical processes using the method of multiple shooting. IET Systems Biology 2007, 1(2):78-88. 10.1049/iet-syb:20060067View ArticleGoogle Scholar
- Ramsay JO, Hooker G, Campbell D, Cao J: Parameter estimation for differential equations: a generalized smoothing approach. Journal of the Royal Statistical Society B 2007, 69(5):741-796. 10.1111/j.1467-9868.2007.00610.xView ArticleMathSciNetGoogle Scholar
- Vilela M, Chou I-C, Vinga S, Vasconcelos ATR, Voit EO, Almeida JS: Parameter optimization in S-system models. BMC Systems Biology 2008, 2, article 35: 1-13.Google Scholar
- Hirsch MW: Systems of differential equations that are competitive or cooperative II: convergence almost everywhere. SIAM Journal on Mathematical Analysis 1985, 16(3):423-439. 10.1137/0516030View ArticleMathSciNetMATHGoogle Scholar
- Smith HL: Systems of ordinary differential equations which generate an order preserving flow. A survey of results. SIAM Review 1988, 30(1):87-113. 10.1137/1030003View ArticleMathSciNetMATHGoogle Scholar
- Schmidt H, Jacobsen EW: Linear systems approach to analysis of complex dynamic behaviours in biochemical networks. IEE Proceedings Systems Biology 2004, 1(1):149-158.View ArticleGoogle Scholar
- Guckenheimer J, Holmes P: Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, Applied Mathematical Sciences. Volume 42. Springer, New York, NY, USA; 1990.Google Scholar
- Perko L: Differential Equations and Dynamical Systems, Texts in Applied Mathematics. Springer, New York, NY, USA; 1991.View ArticleGoogle Scholar
- Gu K, Kharitonov VL, Chen J: Stability of Time-Delay Systems. BirkhÃ¤user, Boston, Mass, USA; 2003.View ArticleMATHGoogle Scholar
- Gopalsamy K: Stability and Oscillations in Delay Differential Equations of Population Dynamics, Mathematics and Its Applications. Volume 74. Kluwer Academic Publishers, Dordrecht, The Netherlands; 1992.View ArticleGoogle Scholar
- Kuang Y: Delay Differential Equations: With Applications in Population Dynamics, Mathematics in Science and Engineering. Volume 191. Academic Press, San Diego, Calif, USA; 1993.Google Scholar
- Jarlebring E: Critical delays and polynomial eigenvalue problems. Journal of Computational and Applied Mathematics. In press
- Murray JD: Mathematical Biology: An Introduction, Interdisciplinary Applied Mathematics. Volume 17. Springer, New York, NY, USA; 2002.Google Scholar
- Mallet-Paret J, Smith HL: The PoincarÃ©-Bendixson theorem for monotone cyclic feedback systems. Journal of Dynamics and Differential Equations 1990, 2(4):367-421. 10.1007/BF01054041View ArticleMathSciNetMATHGoogle Scholar
- Mallet-Paret J, Sell GR: The PoincarÃ©-Bendixson theorem for monotone cyclic feedback systems with delay. Journal of Differential Equations 1996, 125(2):441-489. 10.1006/jdeq.1996.0037View ArticleMathSciNetMATHGoogle Scholar
- Mees AI, Rapp PE: Periodic metabolic systems: oscillations in multiple-loop negative feedback biochemical control networks. Journal of Mathematical Biology 1978, 5(2):99-114. 10.1007/BF00275893View ArticleMathSciNetMATHGoogle Scholar
- Ermentrout B: Simulating, Analyzing, and Animating Dynamical Systems: A Guide to XPPAUT for Researchers and Students. SIAM, Philadelphia, Pa, USA; 2002.View ArticleGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.