- freely available
Cells 2013, 2(4), 635-688; doi:10.3390/cells2040635
Abstract: Systems biology aims at achieving a system-level understanding of living organisms and applying this knowledge to various fields such as synthetic biology, metabolic engineering, and medicine. System-level understanding of living organisms can be derived from insight into: (i) system structure and the mechanism of biological networks such as gene regulation, protein interactions, signaling, and metabolic pathways; (ii) system dynamics of biological networks, which provides an understanding of stability, robustness, and transduction ability through system identification, and through system analysis methods; (iii) system control methods at different levels of biological networks, which provide an understanding of systematic mechanisms to robustly control system states, minimize malfunctions, and provide potential therapeutic targets in disease treatment; (iv) systematic design methods for the modification and construction of biological networks with desired behaviors, which provide system design principles and system simulations for synthetic biology designs and systems metabolic engineering. This review describes current developments in systems biology, systems synthetic biology, and systems metabolic engineering for engineering and biology researchers. We also discuss challenges and future prospects for systems biology and the concept of systems biology as an integrated platform for bioinformatics, systems synthetic biology, and systems metabolic engineering.
The Human Genome Project and high-throughput experimental methodologies such as microarray chromatin-immunoprecipitation DNA chips (ChIP-chip) have led to the development of biology as an increasingly information-rich science encompassing transcriptomes, proteomes, metabolomes, interactomes, and so forth [1,2]. Some have suggested that systems biology is nothing more than a new name for integrative physiology, which has been practiced for the past 50 years or more. Because of these novel technologies, biologists have been able to collect data at a rate that was unimaginable a decade ago. The context of biology has profoundly changed over the past 20 years. These changes provide a powerful new framework for systems biology that moves it far beyond classical integrative physiology. A systems biology approach implies that every system of any level of biological systems can be analyzed with respect to the system’s structure, in particular, in terms of its dynamics, method of control, and method of system design. Systems biology involves genomic, transcriptomic, proteomic, and metabolic investigations from a systematic perspective. As a result, systems biology has become the frontier of modern biological research; large amounts of new omics data cannot be understood without a network or systems viewpoint and without highly sophisticated computational analyses [3,4,5,6,7,8,9,10,11].
The role of systems biology in modern biological research (Figure 1) requires powerful computational tools to mine large-scale data sets of information on genetics, proteins, DNA–protein binding, metabolism, and so forth. These tools are used to construct dynamic system models for the interpretation of specific mechanisms of some cellular phenotypse (or behaviors) from a system (or network) perspective [12,13,14,15,16,17]. To construct a dynamic system model of biological networks from omics data, system identification technologies (i.e., reverse-engineering schemes) are needed to estimate the parameter values of dynamic models and the order of biological networks [18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44]. Synthetic biology metabolic engineering has recently been developed for designing synthetic genetic networks for the production of specific cellular functions in host cells [45,46,47,48,49,50,51]. Based on system models and mechanisms in systems biology, synthetic genetic circuits and metabolic engineering pathways can be designed to investigate cellular behaviors [52,53,54,55,56,57,58,59,60,61,62,63]. These synthetic biological technologies can be employed to investigate the models and mechanisms of systems biology. Discrepancies between the real behavior of synthetic genetic networks and the desired behavior predicted by the models and mechanisms of systems biology can be fed back to modify the models through methodologies of systems synthetic biology and systems metabolic engineering. Based on the role of systems biology (Figure 1), this review describes current developments in bioinformatics, systems synthetic biology, and systems metabolic engineering. It discusses how systems biology can serve as an integrated platform for bioinformatics, systems synthetic biology, and systems metabolic engineering in the future.
Bioinformatics is crucial in genome-wide analyses for understanding cell physiology at different cellular levels (i.e., genome, transcriptome, proteome, and metabolome levels) [12,13]. The various disciplines of bioinformatics provide invaluable information on the global cellular status for systems biology, systems synthetic biology, and systems metabolic engineering, as well as a thorough analysis of the cell. Genomic information in bioinformatics represents the whole genetic makeup of the organism [12,13,14,15,16,17], and comparative genomic analysis may contribute to systems synthetic biology or systems metabolic biology for targeting and engineering genetic circuits to create desirable cellular phenotypes. Transcriptome profiling uses DNA microarrays to decipher the expression levels of thousands of genes under various biological conditions [14,15]. The results can be used to select candidate genes for modification based on systematic analysis of regulatory genes in response to genetic variations and environmental changes, or to identify novel factors for the enhancement of heterologous product secretion in metabolic pathways [64,65,66,67,68]. Proteome profiling is also useful in obtaining transcriptome-profiling data at the protein level. The metabolome comprises the entirety of information on metabolites present within and/or outside the cell under specified conditions [61,62,63]. It is expected to contribute significantly to the understanding of the cell and of synthetic circuit engineering in its metabolic pathways. In this paper, recent advances in the application of bioinformatics to systems synthetic biology and systems metabolic engineering through systems biology are reviewed using specific examples.
Although bioinformatic information (e.g., data on genetics, protein binding, and metabolism) is available, several stages of systems biology are required to help us understand via system dynamics modeling the underlying molecular mechanisms of genetic regulatory (GR) networks [18,19,20,21,22,23,24], protein–protein interaction (PPI) networks [25,26,27,28,29], and metabolic networks [61,62,63] under various biological conditions. At the first stage, a putative GR or PPI network is created by large-scale integration of knowledge such as information from publications and databases, and high-throughput data (from data mining or deep curation). Based on this network and dynamic modeling, the actual GR or PPI network of cellular physiology can be identified with system identification methods (reverse-engineering scheme ) by using specific microarray gene or protein expression data [18,19,20,21,22,23,24,25,26,27,28]. For example, GR networks (GRNs) have been constructed by dynamic modeling via microarray data for cell cycles [18,23,24], environmental stressors [28,44], photosynthesis , aging , and cancer . PPI networks have been constructed for cancer [3,9,33,39,70], inflammation , biofilm formation , and infection by Candida albicans. Comparison of PPI networks between healthy and cancer cells can provide network markers for the investigation of the systematic mechanism of cancer . The integration of cellular networks of GRs and PPIs provides deeper insight into actual biological networks and is more predictive than an approach without integration . A systematic and efficient method to integrate different kinds of omics data for the construction of integrated cellular networks via microarray data have been provided based on coupling dynamic models and statistical assessments . This method has been shown to be powerful and flexible, and can be used to construct integrated networks at different cellular levels to investigate cellular machinery under different biological conditions and for different species. Coupling dynamic models of the whole integrated cellular network is very useful for theoretical analyses and for further experiments in the field of network biology and synthetic biology.
In short, synthetic biology is the engineering of biological systems to fulfill a particular purpose. It is possible to build living machines from off-the-shelf genetic devices by employing many of the same strategies that electrical engineers use to manufacture computer chips [47,48,49]. The main goal of this nascent field is the design and construction of biological systems with desired behaviors . Synthetic biology envisions the redesign of natural biological systems as well as the construction of functional “genetic circuits” by using a set of powerful biotechniques for the automated synthesis of DNA molecules and their assembly into genes and microbial genomes . Synthetic biology is predicted to have important applications in biotechnology, metabolic engineering, and medicine. It may revolutionize how we conceptualize and approach the engineering of biological systems . As illustrated in Figure 1, synthetic genetic circuits can furthermore be employed to confirm network mechanisms derived using systems biology methods, and can be used as feedback for their improvement or revision. Synthetic biology is therefore expected to contribute significantly to a better understanding of the functioning of complex biological systems such as metabolic pathways.
However, the development of synthetic gene networks is still difficult. Most newly created gene networks are nonfunctional because of intrinsic parameter fluctuations, environmental disturbances, and functional variations in the intra- and extracellular context. For this reason, the design method based on dynamic models for robust synthetic gene networks has become an important topic in synthetic biology [52,53,54,55,56,57,58,59,60,68]. These system-dynamics-based design methods for synthetic biology lead to systems synthetic biology.
Heterologous genes have previously been combined into pathways in metabolic engineering, generating a myriad of non-native biochemical products, including isoprenoids, hydroxyacids, biofuels, polypeptides, and biopolymers [64,65,66,67]. Synthetic biologists developed synthetic tools to engineer genetic devices capable of performing complex biological functions such as sensing cell states, counting cellular events, and implementing computational logic. These tools have been applied to the modification and control of metabolic pathways in several organisms. They consist of one or more parts that have been combined to perform a complex function, and provide metabolic engineers with novel ways of exerting cellular control over heterologous production pathways. Some synthetic biological devices with potential relevance in metabolic engineering include orthogonal inducible promoters, light-sensitive promoters, state sensors, spatiotemporal controllers, and logic gates, as well as promoter and ribosome binding site (RBS) libraries [36,37]. Since metabolic engineering seeks to control cellular metabolism and manipulate through heterologous pathways to maximize production of a desired molecule, metabolic engineers are need elegant methods for gathering bio-information about cells, their environment, and modulating gene expression in responses [37,61,62,63]. Hence, devices of synthetic biology promise to be a useful addition to the metabolic engineering toolbox. Some synthetic devices have already been used to increase product titers. However, many remain largely untested in an industrial setting, and the complexity of biology makes their application a feat of engineering [36,37,72]. From the systems biology perspective, continuous work with these devices can help elucidate design rules or aid the development of system dynamics models that facilitate their integration into metabolic industrial processes [36,37,72], and thereby lead to the development of systems metabolic engineering.
Several mathematical techniques based on systems biology have been developed to analyze the systematic properties of complex biological networks. For example, system sensitivity of a biological network in response to various parameter variations has been analyzed to determine the systematic properties that affect the robustness and fragility of a biological network. System sensitivity analysis not only can reveal the robust stability of a biological network against various perturbations, but may also provide information about the controllability of a biological network [7,8,9]. The system response ability of a biological network is a measure of response to environmental signals or disturbances [28,34]. From the system theory perspective, robustness to intrinsic system variation and the ability to respond to external stimuli are two important and complementary system characteristics in the evaluation of system performance [33,34]. A more biological system that is robust toward a large amount of intrinsic parameter fluctuations is less responsive to environmental disturbances, and vice versa. A systems biology investigation of the aging-related gene network via microarray data found that network robustness increases and network response ability decreases during the aging process . The sensitivity of a biological genetic system to environmental molecular noise is considered as an indication of the noise-filtering ability of the gene network . Systems biology allows the measurement of its system characteristics, as well as the capabilities of the signal transduction pathway  Similarly, flux amplification of the metabolic pathway can be estimated, by using both system dynamics models.
As nonlinear biological networks operate under different conditions of cellular homeostasis and homeodynamics, systems biology studies on complex biological models in the landscape of phenotypes are highly informative. These studies help discover possible equilibrium points (phenotypes) and dynamic behaviors, such as bifurcation, oscillation, robust stability, and phase drift to other equilibrium points (phenotype transition). Bifurcation analysis and phase-plane analysis of nonlinear dynamic networks can be useful in predicting system behavior of biological networks under intrinsic parameter changes. Through systems biology approach and dynamic modeling [38,39,40,41,42], network robustness and noise filtering ability can be improved via feedback, redundancy, and modular schemes. This is why there are so many feedback loops, redundant genes, and modular structures at different scales of biological networks. A unifying mathematical framework based on nonlinear stochastic dynamic models [73,74,75] was recently proposed to describe different levels of stochastic biological networks under different parameter fluctuations, genetic variations, and environmental disturbances [29,30,59]. The phenotype robustness criteria of biological networks in systems, evolutionary, ecological, and synthetic biology were investigated from a systematic perspective on the basis of robust stabilization and filtering ability. Network robustness of biological networks can confer intrinsic robustness toward intrinsic parameter fluctuations, genetic robustness for buffering genetic variations, and environmental robustness for resisting environmental disturbances. It was found that if the sum of intrinsic robustness, genetic robustness, and environmental robustness is less than or equal to the network robustness, then the phenotype is robust in different levels of biological networks in systems, evolutionary, ecological, synthetic, and metabolic biology. These phenotype criteria at different levels of biological networks are useful for the design of synthetic and metabolic systems. A systems biology approach based on dynamic models can clearly provide not only a systematic insight into behaviors at different levels of biological networks, but also a design platform to improve system robustness, filtering ability, and transduction ability of synthetic and metabolic system networks, which are discussed in detail in the following sections.
2. Systems Biology Approach to GRNs and PPI Networks via Bioinformatics Methodology
2.1. Construction of GRNs via Microarray Data
The construction of a GRN using a systems biology approach can be divided into two steps. The first step is construction of a potential (or putative) GRN from two high-throughput data sets, namely, a ChIp network and a mutant network. The ChIp data are based on experimental data published by Harbison et al. , who also used a genomic tiling array to identify the genomic region bound by transcription factors (TFs). The mutant data are the gene expression data matrix published by Hughes et al.  with different gene deletion mutants. In general, the potential GRN is suitable for all possible biological conditions. Therefore, the GRN for a specific biological condition needs to be confirmed using microarray gene expression data of the specific biological condition; that is, the real GRN is derived by pruning the potential GRN with specific microarray data. Let xi(t) denote the gene expression of the i-th target gene in the potential GRN. The dynamic equation for the regulation of the i-th gene is then modeled as [21,22]
The regulatory parameter can be estimated from microarray data of the corresponding gene through the least-squares or maximum-likelihood parameter estimation  (known as the reverse-engineering method). After regulator parameters in θi are estimated, the system order (i.e., the number of regulatory genes) is determined by model comparison (e.g., Akaike’s information criterion (AIC)), and the P-value statistical method is employed to determine the significant regulatory genes for target gene i. This is done by pruning false-positive regulations in the potential GRN. That is, some aij is pruned because of false positive deletion. Based on the dynamic model in (2.1), the true GRN can then be constructed one target gene at a time through microarray data. Using similar methods, GRNs for yeast cell cycles [18,23,24], cancer cell cycles , stress response , and inflammation  can be constructed.
2.2. Construction of PPI Networks
The construction of PPI network via a systems biology approach is also a two-step process. The first step is constructing a potential PPI network via data mining from literature and databases such as BioGRID, SGD, and GO [16,17]. As this is only a candidate network based on many biological conditions, the second step is pruning false positive PPIs by using a protein expression microarray of a specific biological condition. For a target protein i in the potential PPI network, the dynamic model of protein activity is [19,20]
The interaction parameter θi can be estimated from protein profile microarray data by least-squares or maximum-likelihood parameter estimation  (if protein profile microarray data are unavailable, ten mRNA microarray data could be used to replace them, with some modification [19,20]). By using AIC to prune false positive interactions, the real PPI network can then be constructed one target protein at a time by following the above two-step procedure. Some dynamic metabolic pathways  and PPI networks of cancer  and inflammation  have recently been constructed by using the microarray data and AIC method. Comparison of PPI networks between healthy and cancer cells can provide network-based biomarkers for molecular investigation and diagnosis of cancer .
2.3. Construction of Integrated GRN and PPI Cellular Networks
Living organisms have evolved complex mechanisms to respond to changes in environmental conditions. This is the case even in unicellular organisms like the yeast Saccharomyces cerevisiae. Such environmental changes, commonly termed as “stress”, are harmful or even lethal to the survival of the cells, especially of microorganisms whose environment is very changeable. Cellular responses to stresses require complex cellular networks of sensing and signal transduction pathways, as well as metabolic pathways to adapt cell growth and proliferation for adjustments of gene expression programs, metabolic activities, and other features of the cell (see Figure 2).
These regulatory systems and signal and metabolic pathways can therefore be suitably described by an integrated cellular network of transcription regulation and PPIs. The flowchart for constructing the integrated cellular network is shown in Figure S1. The same two steps already described apply (construction of a potential cellular network and pruning of false positives via specific microarrays of gene expression and protein expression data). Several kinds of omics data and database information were integrated, including data on microarray gene expression and protein expression, regulatory associations between TFs and genes, and PPI. From these data, the potential integrated GRN and PPI networks are retrieved. In the second step, based on the dynamic model of integrated gene/protein interactions in the potential cellular network, we find that 
Interaction and coupling among genes and proteins in the integrated cellular network in (2.5) are described as follows. If some TFs yj(t) at the end of the signal regulatory pathway regulate their target genes through the regulation function zj(t) = fj(yj(t)) in (2.6), then the regulatory genes influence their corresponding proteins in signal and metabolic pathways through translation effect αnxn(t). The interplay between genes and proteins can be seen from their coupling dynamic equations in (2.5) and Figure 2. Here, the TFs serve as the interface between the signaling regulatory pathway and gene regulatory network. In other words, the interplay between transcription regulation and PPIs constitutes the framework of the integrated cellular network.
Based on the dynamic coupling equation in (2.5), the potential integrated cellular network can be linked through the regulatory parameter aij between genes and their possible regulatory TFs and through the translation parameter αn for gene expression to protein expression. The potential signaling or metabolic pathways can be linked through the interaction parameter bnm between possible interaction proteins. Since omics data on the potential gene regulatory network and potential signaling or metabolic pathway only indicate possible TF-gene regulation and protein interactions, they should be confirmed using microarray data of gene and protein expressions. In particular, values of aij and bnm in (2.5) should be identified and validated by least-squares estimation via microarray data in a specific biological condition or phenotype . Significant regulations and interactions between genes and proteins were detected using model selection methods such as AIC and statistical assessments like such as Student’s t-test . Based on the interface between gene regulatory and signal/metabolic networks (i.e., transcription factors) in a specific biological condition, the two networks are coupled together to form the integrated cellular network. The integrated cellular network for S. cerevisiae under hyperosmotic stress is shown in Figure 3. After the construction of GRN and PPI network from microarray data and bioinformatic method, various system characteristics of the biological network are estimated or measured using systems biology methods in the following sections so that these systematic methodologies can be applied the systematic design of systems synthetic biology and systems metabolic engineering.
2.4. Network Robustness and Sensitivity Estimation via Microarray Data Using a Systems Biology Approach
After the GRNs or PPI networks or their integrated cellular networks are constructed by dynamic modeling using microarray data, some characteristics of these biological networks need to be estimated to gain insight into their systematic mechanisms. “Robustness”, the network ability to maintain systematic performance under intrinsic perturbations, and “response ability”, the network sensitivity to respond to external stimuli or to transduce them to downstream regulators, are two important complementary or antagonistic system characteristics that must be considered when discussing biological network performance. However, these systematic features cannot be measured directly for all network components in the experimental procedure. Even biological processes such as development, differentiation, tumorigenesis, and aging are increasingly being described in terms of temporal ordering of highly orchestrated transcriptional programs [33,34]. The term robustness is encountered widely in diverse scientific fields, from engineering and control theory [73,74,75] to dynamic systems  and biology . It is important to note that robustness describes a relative property, not an absolute one, as no system can maintain stability in all functions when it is perturbed. In other words, robustness is not immutable.
To simplify analysis of “robustness” of the steady state (phenotype), the origin of the dynamic system is shifted to the steady state xs, that is, x(t) = (t) + xs. This shift allows the following shifted dynamic system to be achieved by subtracting Equation (2.7) from Equation (2.8) :
Therefore, the robustness of the phenotype (i.e., the steady state) of the gene network becomes the robustness of the shifted gene network in Equation (2.9) at the origin (t) ≡ 0. In the following, its network robustness (tolerance to intrinsic network perturbation) is tested. Suppose the linear GRN suffers from intrinsic molecular perturbations mainly due to process noise, thermal fluctuation, or genetic mutations. The interactive matrix Â is then perturbed as Â(1 + η), where η denotes the ratio of intrinsic perturbation. That is, the corresponding additional system perturbation is ΔÂ = ηÂ . A GRN with intrinsic perturbation can then be represented by
Because quadratic stability with the Lyapunov energy-like function V( ) = T(t)P (t) > 0, with V(0) = 0 for a positive symmetric matrix, P = PT > 0. The perturbative GRN in Equation (2.10) is robustly stable if ΔV( ) = V( (t+1)) − V( (t)) ≤ 0, i.e., the energy of the gene network is not increased by intrinsic perturbations . Based on this idea of robust stability, if the following inequality has a positive definite solution P = PT > 0 ,
That is, the network robustness is the maximum perturbation ratio η° tolerated by the GRN such that network stability (or phenotype) is still maintained. The constrained optimization problem in (2.12) can be solved by increasing η until no positive solution P exists in Equation (2.11) up to the highest value possible without violating the robust stability in Equation (2.11). A positive definite solution P > 0 in Equation (2.11) can be easily obtained by using the linear matrix inequality (LMI) Toolbox of Matlab. This network robustness method has been used to measure the relative network robustness of multiple loops of a gene regulatory network associated with aging-related pathophysiological phenotypes by using previously reported microarray data. It profiles the effects of aging on gene expression in the thymus, spinal cord, and eye tissues in mice . This aging-related GRN includes 16 genes (Figure 4). The relative robustness η° toward normalized perturbation for GRNs of young, aged, and calorie-restriction (CR) groups is shown in Table 1.
After the network robustness of aging-related GRNs is measured, the response abilities of genes to external stimuli are examined. Assuming that the GRN responds to external stimuli U(t), including upstream regulatory signals and external signals outside the network (e.g., hormones, carcinogens, oxidative stress, or ambient biomedical molecules) that could produce an effect on the gene network, the dynamic Equation (2.9) could be modified as follows:
The effect of input signals U(t) on output signal Y(t) is less than or equal to a positive value δ, if the following inequality holds 
That is, if the above LMI holds for some P > 0, then the effect of U(t) on Y(t) must be less than or equal to δ, and Equation (2.14) holds. The response ability (minimal δ) of the GRN to external stimuli can be obtained by solving the following constrained optimization problem:
This can be solved by decreasing the upper bound δ in Equation (2.14) until no P > 0 exists in Equation (2.15) by using Matlab LMI Toolbox. By following Equation (2.16), the network response abilities of aging-related GRNs in Figure 4 can be measured and compared at different life stages. Results are shown in Table 1.
Table 1 shows that the “aged” group has higher network robustness but lower network response ability. In order to tolerate the intrinsic parameter fluctuations accumulated due to genetic mutation, the elderly GRN becomes more robust than the young GRN. Robust GRNs are less responsive to external stimuli; consequently, the network protective ability against external stimuli decreases in the older gene networks. However, a more robust GRN may harbor more accumulated genetic mutations, which through random drift might provide more evolutionary paths to other phenotypes of gene network and thus lead to some aging-related chronic diseases like cancers, metabolic disorders, and arthritis. Table 1 also shows that the “young” network groups are less robust, with greater response to external stimuli. These GRNs are therefore also less robust to intrinsic perturbations and elicit a strong response toward external stimuli. This might imply that some gene expressions such networks could be easily reprogrammed to mediate downstream genes or regulators for further reactions. It also allows for modulation of gene expression in response to external stimuli, such as exposure to oxidative stress, carcinogens, and pro-inflammation molecules. These observations suggest that the gene regulatory network at the earlier stage of life or under conditions of CR may opportunely have protective adaptations to maintain intact regulatory structures and homeostasis of cellular functions. The purpose of the above analysis is to gain a greater understanding of the systematic protective and/or defensive mechanisms inherent to aging-related GRNs. The proposed robustness measurement methods may be used for future studies of GRNs involved in various biological processes and may have potential applications in gene therapy and drug target selection. It was also found that the network robustness of cancer cells is higher than that of normal cells and that the reverse is true for network response ability [33,34].
2.5. Network-Based Biomarker Determination via Sample Microarray Data Using a Systems Biology Approach
Biomarkers are used either in diagnostic evaluation to determine the health of a patient with or without a disease, or as a prognostic indicator to determine a patient’s prognosis. Present biomarker identification methods, which strictly use gene expression profiles, cannot show how the different genes within the biomarker gene set are related to each other; that is, biomarkers are not identified from a systems biology perspective. Furthermore, the gene lists obtained for similarly diagnosed patients by different research groups differ widely and share few common genes. Here, a systems biology approach is introduced for the integration of microarray data and PPI information to develop a network-based biomarker for systematic investigation into the network mechanism of lung carcinogenesis and the diagnosis of lung cancer . The network-based biomarker consists of two protein association networks constructed from cancer and noncancer samples. The proposed method can be widely applied to determining network-based biomarkers for other diseases. The overall flowchart of the proposed network-based biomarker approach is shown in Figure S2 . First, PPI information and microarray sample data for smokers with and without cancer are used to construct potential PPI networks for cancer and noncancer samples. Since data for both samples are limited, the number of proteins used in potential PPI network construction is also restricted: to avoid overfitting in network construction, the maximum degree of proteins in the potential network are less than the number of samples, imposing a limit on the size of the potential network.
Using a simple regression model, the potential PPI network is then further validated by the sample microarray data to highlight independent protein associations of both samples relative to their respective data sets. For a target protein i in the potential PPI network, the protein is described using the following protein association model :
The associated interaction parameter αik in Equation (2.17) is identified through maximum likelihood estimation  on microarray data. AIC and Student’s t-test were employed for model order selection and for tests on the statistical significance of protein associations. Based on αik, two matrices are established to represent the cancer protein association network (CPAN) and the non-cancer protein association network (NPAN) as follows 
For the i-th protein in the network-based biomarker, the implication of Equation (2.21) is that the CRV quantifies the extent of protein associations that differentiate CPAN from NPAN.
The above network-based biomarker approach is applied to the molecular investigation and diagnosis of lung cancer. The primary data set of GSE4115 (79 smokers with lung cancer and 73 smokers without lung cancer; obtained from the GEO database, http://www.ncbi.nlm.nih.gov/geo/) was used for construction of the network-based biomarker. The CPAN and NPAN in Equation (2.18), which consist of 399 and 393 protein associations respectively, constitute the network-based biomarker of lung cancer (Figure 5). The difference between CPAN and NPAN is shown in Figure 6. The CPAN indicates that 40 identified proteins play significant roles in lung carcinogenesis (Table S1).
2.6. On the Network Robustness and Filtering Ability versus Molecular Noise in GRNs Using a Stochastic System Approach
Molecular noise has been shown to play many roles in the cellular functions of GRNs, including noise-driven divergence of cell fates and population heterogeneity, noise-induced amplification of signals, generation of errors in DNA replication leading to mutation and evolution, and maintenance of the quantitative individuality of cells . Other cellular processes influenced by noise include ion-channel gating, neural firing, developmental modules, cytoskeleton dynamics, and motors . Phase variation in pathogenic bacteria, in which cells alternate randomly between expressing certain genes and silencing others, is thought to be a form of cultivated noise . These molecular-level noisy phenomena are deeply rooted in the statistical mechanical behavior of so-called nanoscale chemical systems, where concentrations of reactive molecule species are extremely low and, consequently, fluctuations (noises) in the reaction rates are large . Even though the molecular fluctuations leading to phase variation seem random in the individual, regulatory factors tune the variation to ensure mean levels of heterogeneity for the population, i.e., the random molecular noises can be shown to be filtered or attenuated by the GRNs .
Since cellular molecular events are subject to significant thermal fluctuations and noisy processes with transcriptional control, alternative splicing, translation, diffusion, and chemical modification reactions, gene expression is best viewed as a stochastic process. Many observations suggest that molecular events underlying cellular physiology are subject to random fluctuations; these observations have led to the proposal of a stochastic model for gene expression and cellular functions . Noise filtering can be considered from a signal processing perspective . From this perspective, a pathway is viewed as an analog filter in terms of its frequency response. In terms of signal processing, these cellular pathways function as biological filters that transduce molecular signals and filter molecular noise .
Suppose the linear GRN suffers intrinsic molecular fluctuations so that the stoichiometric matrix N is perturbed as N + ΔN,
The GRN with random parametric fluctuation can then be represented by
The stochastic dynamic equations of GRNs are always nonlinear. In order to meet the nonlinear stochastic regulatory networks, Equation (2.23) should be generalized as the following Langevin equation 
Based on stochastic Lyapunov theory, let V(x) > 0 with V(0) = 0 denote the Lyapunov power-like function, then the stochastic GRN in Equation (2.23) or (2.24) is stochastically stable if E[dV(x(t))/dt] ≤ 0 . With the choice of V(x) = xT(t)Px(t) for some positive definite matrix P, the following result is derived.
Proposition 1 :
The linear GRN with stochastic perturbation in Equation (2.23) or (2.24) is stochastically stable if the following LMI
Remark 1: (i) In the intrinsic noise-free case, the stable condition in (2.26) is reduced to PN + NTP ≤ 0, i.e., the eigenvalues of system matrix N should be on the left-hand side of the complex domain. Obviously, if the LMI in Equation (2.26) has a positive solution P > 0, then the eigenvalues of N should be located on the far left-hand side of the complex domain with more negative real values in order to overcome the additional term MTPM due to intrinsic random noise. (ii) If some eigenvalues of system interaction matrix N are near the jω axis, then intrinsic random molecular fluctuations across the jω axis perturb these modes more easily such that the linear GRN becomes unstable. The LMI in Equation (2.26) can be rearranged to
For the nonlinear stochastic GRN in Equation (2.25), the following result is also derived.
Proposition 2 :
The nonlinear stochastic GRN in Equation (2.25) is still stochastically stable if the following Hamilton–Jacobi inequality (HJI) has a positive Lyapunov solution V(x) > 0 with V(0) = 0
After the robust stability of GRN is guaranteed under intrinsic biochemical stochastic fluctuation, the effect of environmental random molecular noises on the GRN may be discussed. If the linear GRN in Equation (2.24) also suffers from environmental molecular noises ν(t) = [ν1(t) ⋯ vn(t)]T outside the network (see Figure 7), then
From this is derived the following response level result of the GRN in Equation (2.30).
Proposition 3 :
The response level ρ of the linear stochastic GRN in Equation (2.30) is guaranteed if the following matrix inequality has a positive solution P > 0
The optimal response level ρ0 of the linear stochastic GRN (Equation (2.30)) can be obtained by solving the following constrained optimization
Remark 2: (i) If ρ0 < 1, then environmental molecular noises υ(t) are attenuated by the GRN and ρ0 is called the filtering ability of the GRN, i.e., the GRN is less sensitive to environmental noises. If ρ0 > 1, then environmental molecular noises are amplified by the GRN, i.e., the GRN is more sensitive to environmental noises. (ii) Substituting ρ0 for ρ in Equation (2.32) and rearranging, the following phenotype robustness criterion of the stochastic GRN is derived
The phenotype robustness criterion in (2.35) for the stochastic GRN in (2.30) can be denoted as “intrinsic robustness + environmental robustness ≤ network robustness”. In other words, if the sum of intrinsic robustness and environmental robustness is less than network robustness, then the phenotype of the stochastic GRN remains robust under the influence of stochastic intrinsic fluctuation and environmental noises. In order to maintain the phenotype robustness criterion, GRNs need negative feedback loops to improve the network robustness on the right-hand side of Equation (2.35). Parallel loops and modular and redundant structures are also required to reduce the effect of intrinsic parameter variations on the GRN and to resist environmental disturbance, i.e., to provide more intrinsic robustness and environmental robustness on the left hand side of Equation (2.35). This is why feedback loops, parallel loops, and modular and redundant structures are abundant in GRNs as they contribute to phenotype robustness and favor natural selection. (iii) If the network robustness in Equation (2.35) is not large enough to confer intrinsic robustness and environmental robustness simultaneously to maintain the phenotype of the GRN, some negative feedback gene loops are implemented as follows:
It can be seen from the phenotype robustness criterion in Equation (2.37) that an adequate K can improve the network robustness of the stochastic GRN such that it better tolerates intrinsic stochastic fluctuation and is less responsive to environmental noise (i.e. smaller ρ0).
Similarly, the nonlinear stochastic GRN under environmental molecular noises (Equation (2.25)) should be modified as follows:
The phenotype robustness criterion in Equation (2.37) is then modified as follows :
In general, it is very difficult to solve the HJI in Equations (2.29) and (2.39). Global linearization techniques [79,81] or T-S fuzzy methods [82,83] can be employed to interpolate several local linearized systems to approximate the nonlinear stochastic system in Equation (2.38) as follows:
In this situation, the network robustness criterion in Equation (2.39) can be modified accordingly 
From the phenotype robustness criterion in Equation (2.41), it is seen that if local network robustness is large enough for every local linearized GRN to simultaneously provide enough local intrinsic robustness for tolerating local random parameter fluctuations and enough local environmental robustness for resisting local environmental disturbance, then the phenotype of nonlinear stochastic GRN is maintained in spite of intrinsic stochastic parameter fluctuations and environmental molecular noises. If the phenotype robustness criterion in Equation (2.41) cannot be maintained, then negative feedback loops can be engineered to improve network robustness. These feedback loops may have potential applications for some types of therapy and drug target selection.
Systematic methodologies for the analysis of system characteristics of GRNs and PPINs detailed above are useful for the systematic designs of systems synthetic biology discussed in the following section.
3. Systems Synthetic Biology
Synthetic biology can be expected to have important applications in biotechnology and medicine, and to contribute significantly to a better understanding of the functioning of complex biological systems. Synthetic biology is concerned with the engineering of biological systems that fulfill a particular purpose. This is achieved by transformative innovation that makes it possible to build living machines from off-the-shelf chemical ingredients, employing many of the same strategies that electrical engineers use to make computer chips. The main goal of this nascent discipline is the design and construction of biological systems with desired behaviors [45,46,47,48,49]. Synthetic biology envisions the redesign of natural biological systems based on a set of powerful biomolecular techniques for the automated synthesis of DNA molecules and their assembly into genes and microbial genomes, for greater efficiency, as well as for the construction of functional “genetic circuits” and metabolic pathways for practical purposes . The construction of GRNs has recently demonstrated the feasibility of synthetic biology [84,85,86,87]. The design of robust gene networks still presents a difficult challenge, and most newly designed gene networks cannot function properly. Such design failures are mainly due to both intrinsic perturbations such as gene expression noises, splicing, mutation, evolutionary changes, and environmental disturbances such as changing extracellular environments [52,53]. Designing robust synthetic gene networks that can tolerate intrinsic parameter perturbations, attenuate extrinsic disturbances, and function properly in a host cell is therefore an important topic in synthetic biology .
3.1. Design of Specifications-Based Systems Synthetic Biology
Analysis of the dynamic properties of gene networks has been previously implemented using sensitivity analysis, either by qualitative simulation of coarse-grained models or by extensive numerical simulations of nonlinear differential equation models or stochastic dynamic models . For applications in systems synthetic biology, however, these approaches are not satisfactory since local sensitivity analysis can provide only a partial description of all possible behaviors of a nonlinear gene network. In particular, it cannot guarantee that a synthetic network behaves as expected for all uncertain initial conditions, external disturbances, and parameter variations in a given range. Recently, Kuepfer et al. developed an approach based on semidefinite programming for partitioning parameter spaces of polynomial differential equation models into “feasible” and “infeasible” regions . In this approach, “feasible” simply refers to the existence of a desired steady state of the synthetic network. Another approach using robustness analysis and tuning of synthetic networks was proposed by Batt et al. to provide a means to assess the robustness of a synthetic gene network with respect to parameter variations . This approach allows searching for parameter sets for which a given property is satisfied, using a publicly available tool called RoVerGeNe. Several gene circuit design methods have recently been introduced insert or delete specific circuits in an existing gene network to modify its structure toward improved robust stability and filtering ability . Robust synthetic gene network design, however, is a topic in itself, necessitating the design of a completely synthetic network with enough robust stability toward parameter fluctuations and with enough noise filtering ability to resist external disturbances, which allows it to function properly in a host cell.
A robust synthetic biological design incorporating molecular noises has been developed based on stochastic game theory . However, the intrinsic parameter fluctuations of synthetic gene networks have not been considered in the design procedure. In a previous study  some system design specifications (engineering designs) had been provided by users so that the designer must engineer an artificial gene network to meet these design specifications. For the convenience of problem description, a simple design example is provided to give an overview of design problems for robust synthetic gene networks. A more general problem scenario is treated in preparation for this overview. First, consider the cross-inhibition network shown in Figure 8. This network is synthesized with two genes (a, b) that code for two repressor proteins (A, B). More specifically, protein A represses the expression of gene b, and, at higher concentration, the expression of its own gene. Protein degradation is not regulated. This synthetic system can be modeled by the following synthetic equations:
The state variables xa and xb denote the concentrations of proteins A and B, respectively. k and γ are the kinetic parameter and decay rate, respectively. r is the regulation function capturing the regulator effect of a transcriptional protein on gene expression, and has a smooth sigmoidal form (e.g., Hill function) .
The simple cross-inhibition network in (3.1) and (3.2) can be represented by the following stoichiometric matrix equation 
However, the stoichiometric matrix in vivo suffers from intrinsic parameter perturbations because of gene expression noises, splicing, mutation, evolutionary change, etc., as in 
Suppose the synthetic gene network also suffers from environmental disturbances because of changing extracellular environments and interactions with the cellular context in its host cell. The stochastic gene network can be then represented as
If a synthetic gene network consists of n genes, then the stochastic gene network of Equation (3.5) in the host cell can be extended according to the following n-gene network dynamics 
To ensure correct and efficient operation of the gene network, several systematic design specifications should be imposed on it from the systematic engineering point of view.
(i) The kinetic parameters and the decay rates in stoichiometric matrix should be chosen from a biologically feasible range
(ii) The network should tolerate the stochastic kinetic parameter and decay rate fluctuations with prescribed standard deviations in Mi in the state-dependent noise term
(iii) The following desired steady state should be achieved
(iv) The following prescribed attenuation level ρ of environmental disturbance should be achieved (i.e., the H∞ filtering ability)
In order to achieve the desired steady state xd and for convenience of design, the origin of the nonlinear stochastic system in Equation (3.7) should be shifted to xd. Stabilizing the shifted nonlinear stochastic system at the origin would then also achieve xd, simplifying the design procedure. Let = x – xd. The following shifted stochastic synthetic genetic system is then derived :
From the stochastic synthetic gene network Equation (3.9) in vivo, we obtain the following result.
Proposition 4 :
If design kinetic parameters and decay rates in N ∈ [N1,N2] are chosen such that the following HJI has a positive solution V( ) > 0
It is generally very difficult to specify N ∈ [N1,N2] to solve HJI in Equation (3.11) for V( ) > 0 using a systematic method. If all global linearizations are bound by a polytope consisting of M vertices
Proposition 5 :
Assuming that design kinetic parameters and decay rates in N ∈ [N1, N2] are chosen such that the following M LMIs have a common symmetric positive definite solution P > 0
Remark 3: (i) Gene circuit design can now be implemented using recombination technology to insert or delete TF binding sites in the promoter region of a regulated gene with the aim of increasing or decreasing the value of the kinetic parameter κi (i.e., different levels of affinity) of the regulated gene . By inserting strong or weak binding sites, large or small values can be obtained. For example, the binding site of κi = 1 is 10 times larger than that of κi = 0.1 at the promoter region of target gene i. Changes to the decay rate γi can be implemented by shortening the 3' polyadenylate tail (referred to as deadenylation), which primarily triggers decapping, resulting in 5' to 3' exonucleolysis. Alternatively, removal of the 3' polyadenylate tail can increase γi . Therefore, by shortening or elongating the gene’s 3' polyadenylate tail, we can increase or decrease the decay rate γi of gene i. Directed evolution methods are also useful in changing the elasticity (kinetic property of κi) and in designing biochemical circuits . From a systems biology perspective, these advances in implementation techniques of κi and γi enable engineering of synthetic gene networks in the near future.
(ii) Because of the uncertain values of v and initial state (0) of the synthetic gene network (3.9) in the host cell, the following minimax design has also been considered to reset the uncertainty of ν(t) and (0) in vivo . ν(t) and (0) are considered players maximizing their effects on regulation error (t) when the kinetic parameters are specified as another player minimizing the effects of v and (0) on (t).
A robust synthetic gene network based on dynamic game theory and fuzzy interpolation of local linearized linear systems can thus be efficiently designed .
(iii) When the desired output of the synthetic gene network is not a steady state but a time-varying signal like a specific binary or periodic signal, and the network structure becomes complex, the above systematic analysis methods are difficult to apply. Because natural selection of traits suited for environmental change is an important evolutionary mechanism, kinetic parameters of synthetic gene networks can be tuned by genetic algorithms (GAs) or evolution algorithms (EAs) to optimally track desired biological functions [54,58]. Based on the evolutionary network algorithm, the kinetic parameters may be tuned to maximize the fitness to some desired phenotype selected by natural selection. Consider the optimal tracking design of a synthetic gene network by network evolution shown in Figure 9. Let the tracking error be defined as
The design purpose is then to tune the design parameter ki using evolutionary (or genetic) algorithms such that the network can achieve the following optimal tracking
Let the fitness function F(k) be defined as
Adapting a parameter vector (chromosome) ki ∈ [kiL,kiU] by EA or GA to minimize J(k) for the desired network behavior tracking of the synthetic gene network is equivalent to maximizing the fitness function in Equation (3.18) to meet the natural selection. A robust biological network design with a desired output behavior yd(t) is therefore equivalent to as solution to the following fitness maximization problem using an evolutionary network method [54,58]
The evolutionary network algorithm (or evolutionary systems biology algorithm ) is employed to solve the above fitness maximization problem via genetic operators such as selection, crossover, and mutation. It mimics natural selection in an evolutionary process to tune the kinetic parameter network of the synthetic gene network to an environmental change. A GA-based algorithm with binary coding of the chromosome has been proposed for the design of robust synthetic genetic oscillators with prescribed amplitude, period, and phase . The oscillator is intended to allow protein concentrations to track desired periodic reference signals under intrinsic and environmental noises. Based on an evolutionary systems biology algorithm that encodes each chromosome in a real-valued vector, the design parameters of target gene circuits can evolve to specific values in order to robustly track a desired biological function in spite of such interferences .
3.2. Robust Synthetic Gene Network Design via Library-Based Search Methods
Over the past decade, synthetic biology has made significant progress in designing biological parts. Even synthetic gene networks may have been constructed using a variety of biological components to achieve desired behaviors. A limitation on the development of complex synthetic gene networks intended to track specific reference trajectories is the lack of an efficient method for selecting suitable biological parts from libraries; experimental data in promoter libraries cannot be directly used for selecting adequate promoter parts. Current promoter libraries therefore need to be redefined based on promoter activity. This would allow development of library-based search methods.
A dynamic model can be used in the indirect evaluation of the activities of promoter parts to help redefine existing promoter libraries. In the TetR-regulated promoter library shown in Figure 10, x(c, t) and X(c, t) respectively denote the concentrations of the mRNA and protein of gene yegfp (which is used to measure protein expression). The dynamic model of the promoter-regulation gene part is constructed with [56,57]
Based on the estimated promoter activities c = (cr, cs) via maximum and minimum values of the steady state of protein expression data, some promoter libraries can be redefined in such a way that they can be efficiently selected from the design of the synthetic gene network (Table 2) [56,57]. Since a synthetic gene network always consists of a set of promoter-regulation gene circuits (Figure 10), the design of complex synthetic gene network addresses how to select a set of promoters from the corresponding promoter libraries that have promoter activities adequate for achieving the design specifications. A well-known gene circuit topology, the simple toggle switch, is shown for illustration purposes in Figure 11. The toggle switch has two distinct stable states and can be reversibly switched between them by changing the inducers ATc and IPTG. Let x1(c1, t), x2(c2, t), and x3(c3, t) denote the concentrations of mRNAs tetR, lacI, and yegfp, respectively; and let X1(c1, t), X2(c2, t), and X3(c3, t) denote the concentrations of proteins TetR, LacI, and yEGFP, respectively. The dynamic model of the toggle switch gene network in Figure 11 is then modeled as
The promoter regulation functions pLacI(c1, r1), pTetR(c2, r2), and pTetR(c3, r3) are dependent on the selection of promoters c1, c2, and c3 from the corresponding promoter libraries in Table 2. The output of interest y(c, t) is dependent of the selected promoter set c = [c1, c2, c3] with adequate promoter activities from the corresponding promoter libraries in Table 2. This dynamic toggle switch gene network model consists of three interactive dynamic models of promoter-regulation gene parts, as shown in Equation (3.20).
|TetR-regulated promoter library (LibTetR)||LacI-regulated promoter library (LibLacI)|
|Promoter||Promoter activity||Promoter||Promoter activity|
The multiobjective design approach for the H2/H∞ synthetic gene network based on promoter libraries selects an adequate promoter set c = [c1, c2, c3] from corresponding promoter libraries such that the following two design objectives are achieved simultaneously :
(i) H∞ desired noise attenuation level ρd:
(ii) H2 optimal reference tracking:
By solving an optimization problem with two constraints , adequate promoters can be selected from the corresponding libraries in Table 2 to achieve the two design objectives in Equations (3.23) and (3.24). Based on the synthetic toggle switch (Figure 11) and dynamic model (equation (3.22)) with intrinsic parameter variation and external disturbance, the adequate promoter set c = [c1, c2, c3] [L9, T2, L8] is selected from promoter libraries in Table 2 to achieve the multi-objective H2/H∞ reference tracking specified in Equations (3.23) and (3.24). Simulation results with ν(t) = 10 × [n1⋯n6]T are shown in Figure 12.
Remark 4: Collective rhythms of GRNs, especially the synchronization of dynamic cells mediated by intercellular communication, have become a subject of considerable interest to biologists and theoreticians . Synchronization of a population of synthetic genetic oscillators is an important consideration in practical applications, because a population distributed over different host cells needs to exploit molecular phenomena in a simultaneous manner in order to function as a biological entity. However, this synchronization of synthetic gene networks in different host cells may be corrupted by intrinsic kinetic parameter fluctuations and extrinsic environmental molecular noise. Therefore, robust synchronization is an important design topic in nonlinear stochastic coupled synthetic genetic oscillators with intrinsic kinetic parametric fluctuations and extrinsic molecular noise. A systems biology approach indicates [59,60] that if the synchronization robustness criterion is greater than or equal to the sum of the intrinsic robustness and extrinsic robustness, then the stochastic coupled synthetic oscillators can be robustly synchronized in spite of intrinsic parameter fluctuation and extrinsic noise. If the criterion for synchronization robustness is violated, then an external control scheme can be designed to improve robustness by adding inducers to the coupled synthetic genetic network. These robust synchronization criteria and control methods are useful for a population of coupled synthetic networks with emergent synchronization behavior, especially for multicellular engineered gene networks .
4. Systems Metabolic Engineering
Systems metabolic engineering aims to amplify or delete specific genes in metabolic pathways to perform metabolic engineering within a systems biology framework. Regulatory gene networks, metabolic networks, and other cellular networks can thus be engineered in an integrated system manner. Systems biological analysis via large-scale genome-wide analyses and computational bioinformatic tools can allow the rapid evaluation of the global physiology of a cell with respect to various cellular regulations. These include transcriptional and translational regulation, as well as metabolic engineering distribution [37,61,62,63]. Results of this systems biology approach can be used to predict targets for a metabolic engineering approach within the host strain. Furthermore, the integration of high-throughput, large-scale, genome-wide analyses with in silico simulation results might provide additional information on cellular status at various hierarchical levels from genome to fluxome. Strain selection for strain improvement by systems metabolic engineering is divided into several phases : (i) A base strain is allowed to develop. (ii) The base strain is further engineered via synthetic biological methods, based on results obtained from high-throughput genome-wide bioinformatic data and systems biology computational analyses. (iii) The performance of this preliminary production strain is evaluated in actual fermentation. (iv) The results are then fed back into further strain development until a superior strain showing the desired performance is derived.
By combining the results of multiple genome-wide analyses and computational analyses, systems biology may allow us to reach an unprecedented level of understanding of cellular physiology and characteristics , which can subsequently be used in a systems synthetic biology framework to design systems metabolic engineering strategies, for example, for strain improvement. However, novel computational methods such as large-scale data mining, multidimensional data integration, and data-driven network inference and deep curation schemes still need to be developed and applied to the integration and interpretation of heterogeneous large-scale genome-wide bioinformatic data, which are closely interconnected by complex regulatory and metabolic pathways [37,91]. Several robust biochemical circuit designs have recently been proposed to improve the network robustness of metabolic pathways. The proposed design schemes provide a systems biology method with potential applications in systems synthetic circuit design in systems metabolic engineering and systems drug design. Broadly, a metabolic network is a collection of enzymatic reactions that process cellular and intercellular metabolites. In systems metabolic engineering, the rates of reactions or fluxes, which correspond directly to changes in concentrations of substrates, enzymes, factors, or products, are often measured. Such systematic changes in concentrations can be expressed in terms of dynamic differential equations. The following S-system representation has been used for the last three decades as an efficient model for describing a dynamic metabolic network [7,27]
Construction of the S-system representation of a metabolic network and estimation of its kinetic parameters from experimental data are described in  and the literature references therein. The nonlinear parameter estimation problem of S-systems has recently been solved efficiently by evolution optimization methods . It is generally difficult to study the network robustness or sensitivity of a nonlinear system such as Equation (4.1). Fortunately, many important characteristics of an S-system at or close to the steady state can be analyzed by using simple algebraic methods. Since most metabolic networks in nature operate near the steady state, at which inputs and outputs are almost balanced, the following focuses on the network robustness of metabolic networks at the steady state [7,27].
4.1. Robust Biochemical Circuit Design in Metabolic Networks
Consider the steady state of the metabolic network in Equation (4.1),
Taking the logarithm on both sides of Equation (4.2) and making some rearrangements,
Introducing new variables and coefficients,
In Equation (4.5), AD denotes the system matrix of the interactions between dependent variables YD, and AI indicates the interactions between dependent variables YD and independent variables YI. In the nominal parameter case, it is assumed that the inverse of AD exists so that YD can be solved uniquely, i.e., the metabolic network results in only one steady state (phenotype). Therefore, the steady state of the biochemical system is given by
Suppose that parameter variations due to mutation, thermal changes, or disease can alter the kinetic properties of the steady state of a metabolic network as follows 
Parameter perturbations are defined as
From Equation (4.7), we derive
If the following robustness condition holds 
The above analysis implies that if the robustness condition in Equation (4.9) holds, then the steady state of a metabolic system is preserved under parameter variations ΔAD, i.e., YD + ΔYD in Equation (4.10) has a small difference ΔYD from the nominal in Equation (4.6) under small perturbation. However, if the condition in Equation (4.9) does not hold, then individual values of I + AD−1ΔAD may be zero, the inverse (I + AD−1ΔAD)−1 may not exist, and the steady state YD + ΔYD may cease to exist under the parameter perturbation ΔAD. As an example, consider the singular value decomposition:
Obviously, the inverse (AD−1ΔAD)−1 or (I+AD−1ΔAD)−1 does not exist under the parameter perturbation in Equation (4.12).
Remark 5: The parameter perturbations in the direction of singular vectors like Equation (4.12) are the network’s points of fragility. The robustness prevents this kind of parameter variation and guarantees the existence of the steady state of the metabolic networks. When unexpected perturbations like Equation (4.12) are encountered, a catastrophic failure of the network follows. Robust circuit design is a necessary fail-safe mechanism in such situation. For example, the trehalose pathway in yeast consists of only a few metabolites that form a substrate cycle. It is governed by a surprisingly complex control system that is composed of several inhibiting or activating signaling mechanisms .
Equivalently, the network robustness of the metabolic network in Equation (4.9) can be rewritten as a more intuitive criterion for network robustness:
That is, ADADT is the upper bound of ΔADΔADT without violation of robust stability at steady state. If the network robustness criterion in Equation (4.14) holds, then the steady state of the perturbative metabolic network still exists.
If a metabolic network cannot tolerate perturbations, i.e., the network robustness criterion in Equation (4.14) is violated, then robust control via a biochemical circuit design is a suitable remedy. Based on robustness analysis, we develop a biochemical circuit design scheme for the robust control of metabolic networks. Consider the robust control system design of the metabolic network (in Equation (4.1)) by specific biochemical feedback circuits in a more general metabolic network form 
Consider the robust control system of the metabolic network in Equation (4.15). By using a similar procedure to Equations (4.2)–(4.7) at steady state,
Suppose we can find some F such that the inverse of (AD + F + ΔAD) exists. Equation (4.16) is equivalent then to
Similar to Equation (4.14), a robust design scheme for the controlled metabolic system in Equation (4.17) is given by
In this case of robust circuit design, the design purpose is to specify feedback circuits such that the structural stability robustness of the metabolic network is improved, thus enabling the network robustness criterion in Equation (4.18) to tolerate larger parameter perturbations ΔAD. The phenotype (i.e., the steady state of the controlled metabolic network in Equation (4.16)) is then given by
A simple example of network robustness analysis and circuit design is given in supplementary example 2.
4.2. Multipurpose Circuit Control Design of Metabolic Networks
The robustness design above focuses on the tolerance of kinetic parameter perturbations ΔAD. The effects of rate constant variations Δb and of environmental changes or upstream regulatory changes ΔYI on output variations ΔYD should also be considered in the circuit design of metabolic networks to guarantee robustness against both intrinsic parameter variations and extrinsic environmental perturbations. The sensitivity of ΔYD to Δb in the designed metabolic network of Equation (4.16) is given by 
The sensitivity of ΔYD to ΔYI is given by
It is more appealing to design a robust metabolic network with desired sensitivities to variations in rate constants and environmental signals, i.e.,
A multipurpose circuit design approach is therefore aimed to construct a biochemical circuit F to simultaneously satisfy the design requirements of network robustness in Equation (4.18) and network sensitivity in Equation (4.23). For example, suppose the goal is to design a robust biochemical circuit f22 in supplementary example 2 to tolerate kinetic parameter variation ΔAD and satisfy the network sensitivity in Equation (4.22) or (4.23) with prescribed sensitivities of ‖ΔYD/Δb‖2 < ‖ΔYD/Δb‖2,nominal ≡ s1 = ‖AD−1‖2 = 3.42 and ‖ΔYD/ΔYI‖2 < ‖ΔYD/ΔYI‖2,nominal ≡ s2 = ‖AD−1AI‖2 = 2.66. Then f22 should be specified to satisfy the robust circuit design and the following inequalities:
Similarly, suppose the goal in the design case in equation is to specify f12 and l22 to satisfy equations and (4.24), i.e.,
Given that the goal is to design a robust metabolic circuit to tolerate the perturbation ΔAD and to achieve the desired sensitivities s1 and s2 in Equation (4.23), the robust control circuit design problem can be reduced to specifying F to satisfy the following multipurpose control circuit design derived from Equations (4.18) and (4.23):
If one biochemical control pathway with kinetic parameter f12 is designed to satisfy the multi-objective design criteria in Equation (4.26), then the range of f12 is found to be within [−0.8, −0.1]. If f12 is chosen as −0.2 (Figure S6(a), blue line), then the time responses of the designed TCA cycle network shown in Figure S6(d), which match the desired properties of the proposed design method, are obtained. That is, the robust controlled biochemical network not only can tolerate ΔAD (to preserve its phenotype under parameter perturbations) but also retains sensitivity to environmental molecules in the nominal case. If dynamic circuit design is employed to implement the biochemical circuit f12, then an enzyme capable of catalyzing the reaction X2→X1 is required. Additionally, a TF (Z) has to be found such that oxaloacetate2 (X2) can bind to the promoter of the enzyme’s inhibitor gene, as f12 is negative. The concentration of X2 could therefore regulate X1 through the kinetic parameter f12. The elasticity of the enzyme inhibitor’s gene sequence has to be modulated then to the specified performance by rational design or directed evolution. This allows the construction of the biochemical control circuit f12.
The previous section discusses robust circuit design of metabolic networks based on the steady state of an S-system model. In the following section, a stochastic dynamic model approach is taken based on the systems biology approach outlined in Section 3.
4.3. Robust Control Circuit Design of Stochastic Dynamic Metabolic Networks
Based on sampled data from biochemical experiments and the delayed effect of molecular diffusion and transport in cells, the linear metabolic regulatory network can be suitably modeled by the following discrete-time dynamic system:
Suppose the kinetic parameters of a biochemical regulatory network of metabolic processes are affected by the following intrinsic perturbation and environmental disturbance:
Based on the robust stabilization and disturbance filtering described in Section 2, the environmental disturbance attenuation level ρ of the metabolic network is denoted as
Following the systems biology approach in Section 2, the following robust result for a linear stochastic metabolic network is derived.
Proposition 6 :
For a linear stochastic metabolic network with intrinsic parameter fluctuation and environmental disturbance, if there exists a positive definite matrix P = PT > 0 such that the following matrix inequality holds for a desired disturbance attenuation level ρ
The phenotype robustness criterion in Equation (4.30) could be rewritten as
Remark 6: (i) The physical interpretation of the phenotype robustness criterion of the metabolic network in Equation (4.31) is that if intrinsic robustness allowing tolerance of intrinsic parameter fluctuation and environmental robustness, as well as attenuation of environmental disturbance are simultaneously conferred by the network robustness, then the phenotype of the metabolic network is maintained. It can be shown that if the eigenvalues of A are closer to the origin, then the network robustness is commensurately larger.
(ii) To solve the phenotype robustness criterion in Equation (4.30), it can be transformed into the following equivalent LMI:
According to its definition, the disturbance attenuation (filtering) ability ρ0 (i.e., the minimum ρ) can be obtained by solving the following constrained optimization problem:
If the metabolic network in Equation (4.28) cannot achieve a prescribed disturbance attenuation ability ρP (i.e., ρP < ρ0) specified for therapeutic or biotechnological purposes, then a biochemical circuit design using state feedback is necessary:
The phenotype robustness criterion in Equation (4.31) is then modified as
The negative feedback circuit F could shift the eigenvalues of A to the origin, thereby improving the network robustness of the metabolic network on the right-hand side of Equation (4.35) and achieving the prescribed attenuation level ρP < ρ0. In general, the interactions of a biochemical regulatory network in metabolic processes are nonlinear in real biosystems. In this situation, a nonlinear biochemical regulatory network of metabolic pathways under intrinsic stochastic parameter perturbation and environmental disturbance can be represented based on the stochastic dynamic model of systems biology in Section 2:
In the following, the robust stability and filtering ability ρ0 on ν(k) at y(k) in the stochastic nonlinear metabolic network in Equation (4.36) is discussed.
Proposition 7 :
If the following matrix inequality holds for a Lyapunov function V(x(k)) > 0 and a noise attenuation level ρ
The filtering ability ρ0 of the nonlinear stochastic metabolic network in Equation (4.36) that attenuates environmental disturbance can be obtained by solving the following constrained optimization problem:
Following the robustness result discussed in Section 2, the phenotype robustness criterion for the nonlinear stochastic metabolic network in Equation (4.36) is given as
Suppose the network robustness of the metabolic network cannot confer enough intrinsic robustness and environmental robustness to maintain the phenotype of the metabolic network. In this situation, some negative feedback loops Fg(X(t)) should be implemented to improve network robustness as follows:
In this case, the phenotype robustness criterion in Equation (4.39) should be modified as follows:
That is, the negative feedback loops Fg(X) are employed to improve network robustness to tolerate more fluctuations of the intrinsic parameter and to filter more environmental disturbances.
It is generally very difficult to solve the HJI in Equations (4.37), (4.39), and (4.41) for robust chemical circuit design of metabolic networks. Based on the global linearization method in Equation (3.12) [79,81] or fuzzy interpolation methods [83,93], the nonlinear stochastic metabolic network can be approximated by interpolating M local metabolic networks as follows:
Suppose the negative feedback loop F is designed for the metabolic network in Equation (4.40) or (4.42), such that the following LMIs have a positive solution P > 0:
If the optimal robust filtering design is employed for the nonlinear stochastic metabolic network in Equation (4.40) or (4.42), then the control matrix F is specified to solve the following constrained optimization problem:
This robust biochemical circuit design example in a metabolic network illustrates that the proposed systematic method of biochemical circuit design produces metabolic networks that not only tolerate more fluctuations of the intrinsic random parameter but also filter more environmental disturbances. Systems metabolic engineering design is thus capable of improving the network robustness of metabolic networks and efficiently attenuating the effect of environmental noises. This approach may serve as a basis for drug designs against genetic perturbations, pathological environmental cues (such as infectious agents or chemical carcinogens), or both.
Network robustness is a systematic property that allows a metabolic network to maintain its biochemical function or to generate biochemical products despite external disturbance and intrinsic parametric fluctuation. It is from a class of fundamental and ubiquitously observed systems-level phenomena that cannot be understood by observing individual components only. Thus, a study of network robustness at one level is simply a study of why, when, and how metabolic networks function properly or otherwise [47,80,94]. If network robustness is not large enough to obtain sufficient intrinsic robustness to tolerate random parameter fluctuations and environmental robustness to filter environmental molecular noises, then a robust biochemical circuit design is necessary to improve robustness such that the network can maintain its function or phenotype. The proposed robust circuit design principles could potentially be used for robust biosynthetic network design with applications in drug design, gene therapy, and metabolic engineering. Future focus may be on robust chemical circuit design for such applications or construction of other pathways by nanotechnology and metabolic engineering . If network robustness, intrinsic robustness, and environmental robustness are taken into consideration in metabolic network engineering, such networks would function more reliably and efficiently.
With the exception of implementing feedback biochemical circuits in metabolic networks, systems synthetic biology techniques detailed in Section 3 could be combined with systems metabolic engineering approaches to engineer complete metabolic pathways or networks to produce biochemical products that could not be produced in the host cell. As an example, E. coli does not have non-fermentative pathways for isobutanol (see Figure 14(A)). Suppose a synthetic pathway in E. coli is required for the production of butanol as biofuel. In this application, it is necessary to engineer transcription and translation genes to produce the enzymes AlsS, IlvC, IlvD, and Kdc (KIVD), which are necessary to catalyze the isobutanol metabolic pathway (Figure 14(B)). The last enzyme of the metabolic pathway (Adh) is already present in E. coli. Promoters and RBSs with different regulatory abilities can be selected from the corresponding libraries. With adequate computation and simulation following the systems biology approach (see Section 2), a synthetic metabolic pathway could be designed and engineered (see Section 3) to achieve high-yield and high-specificity metabolic production of isobutanol.
5. Future Challenges in Systems Biology
Systems biology approaches to synthetic biology and metabolic engineering are built on molecular and genetic findings, and results of studies in omics fields such as genomics, proteomics, and metabolomics. The main concerns faced by systems biology are the complexity and dynamic character of biological systems, the vast quantities of biological data, and the fragmented nature of biological knowledge. These fragments of information need to be integrated with nonlinear stochastic dynamic models at different system levels by suitable computational tools before they can be applied to system synthetic biology and systems metabolic engineering.
The first challenges of systems biology are how to enrich omics data (e.g., updating and improving protein microarray, ChIP-chip databases, and pathway database for systematic studies discussed in chapters 2–4) and develop more powerful computational tools for sophisticated data handling, advanced modeling, integrated analysis, and knowledge integration. A systems biology approach based on the integration of these enriched omics data and advanced computational models will enable us to predict the behaviors of biological systems more precisely. This will increase our understanding of the underlying molecular mechanisms and our ability to efficiently predict the effects of designed genetic circuits in metabolic engineering as well as the impact of perturbations on biological systems in drug treatment.
The second challenge is how to increase the capability of researchers to navigate and relate various data and knowledge resources using the integrated platform of bioinformatics, systems synthetic biology, and system metabolic engineering to enable innovations. Connecting genomics, molecular networks, and physiology will provide us with deeper understanding of how individual differences in the genome affect physiological processes through alterations in molecular networks. Therefore, predictive and preventive medicine based on network-based biomarkers inevitably lead to personalized medicine that may revolutionize healthcare in the future . The third challenge of systems biology is how to integrate bioinformatics, gene circuit design, and metabolic engineering technologies for systems drug design in future predictive and preventative medicine.
Drastically increasing oil consumption, exhaustion of natural resources, and global warming are worldwide concerns at present. They have spurred research in the areas of bioenergy and biorefining, as well as microbial production of bulk chemicals and materials. Systems biology will play an important role in these undertakings, collaborating with systems synthetic biology and systems metabolic engineering to generate bio-based products to replace some, if not most, currently used chemicals and materials. The fourth challenge of systems biology is how to make these bio-based processes competitive.
Supplementary materials can be accessed at: https://www.mdpi.com/2073-4409/2/4/635/S1.
The work was supported by the National Science Council of Taiwan, R.O.C. [NSC 102-2745-E-007-001-ASP].
Conflicts of Interest
The authors declare no conflict of interest.
- Joyce, A.R.; Palsson, B.O. The model organism as a system: integrating 'omics' data sets. Nat. Rev. Mol. Cell. Bio. 2006, 7, 198–210. [Google Scholar] [CrossRef]
- Ideker, T.; Bafna, V.; Lemberger, T. Integrating scientific cultures. Mol. Syst. Biol. 2007, 3, 105–106. [Google Scholar]
- Chuang, H.Y.; Lee, E.; Liu, Y.T.; Lee, D.; Ideker, T. Network-based classification of breast cancer metastasis. Mol. Syst. Biol. 2007, 3, 140–157. [Google Scholar]
- Oti, M.; Snel, B.; Huynen, M.A.; Brunner, H.G. Predicting disease genes using protein-protein interactions. J. Med. Genet. 2006, 43, 691–698. [Google Scholar] [CrossRef]
- Peri, S.; Navarro, J.D.; Amanchy, R.; Kristiansen, T.Z.; Jonnalagadda, C.K.; Surendranath, V.; Niranjan, V.; Muthusamy, B.; Gandhi, T.K.B.; Gronborg, M.; et al. Development of human protein reference database as an initial platform for approaching systems biology in humans. Genome Res. 2003, 13, 2363–2371. [Google Scholar] [CrossRef]
- Zhang, S.H.; Jin, G.X.; Zhang, X.S.; Chen, L.N. Discovering functions and revealing mechanisms at molecular level from biological networks. Proteomics 2007, 7, 2856–2869. [Google Scholar] [CrossRef]
- Voit, E.O. Computational Analysis of Biochemical Systems: a Practical Guide for Biochemists and Molecular Biologists; Cambridge University Press: New York, NY, USA,, 2000. [Google Scholar]
- Chen, B.S.; Wu, C.C. On the calculation of signal transduction ability of signaling transduction pathways in intracellular communication: systematic approach. Bioinformatics 2012, 28, 1604–1611. [Google Scholar] [CrossRef]
- Kitano, H. Opinon—Cancer as a robust system: Implications for anticancer therapy. Nat. Rev. Cancer 2004, 4, 227–235. [Google Scholar] [CrossRef]
- Stelling, J.; Sauer, U.; Szallasi, Z.; Doyle, F.J.; Doyle, J. Robustness of cellular functions. Cell 2004, 118, 675–685. [Google Scholar] [CrossRef]
- Storey, J.D.; Xiao, W.Z.; Leek, J.T.; Tompkins, R.G.; Davis, R.W. Significance analysis of time course microarray experiments. Proc. Natl. Acad. Sci. USA 2005, 102, 12837–12842. [Google Scholar]
- Cherry, J.M.; Adler, C.; Ball, C.; Chervitz, S.A.; Dwight, S.S.; Hester, E.T.; Jia, Y.K.; Juvik, G.; Roe, T.; Schroeder, M.; et al. SGD, Saccharomyces genome database. Nucleic Acids Res. 1998, 26, 73–79. [Google Scholar] [CrossRef]
- Gasch, A.P.; Spellman, P.T.; Kao, C.M.; Carmel-Harel, O.; Eisen, M.B.; Storz, G.; Botstein, D.; Brown, P.O. Genomic expression programs in the response of yeast cells to environmental changes. Mol. Biol. Cell. 2000, 11, 4241–4257. [Google Scholar] [CrossRef]
- Harbison, C.T.; Gordon, D.B.; Lee, T.I.; Rinaldi, N.J.; Macisaac, K.D.; Danford, T.W.; Hannett, N.M.; Tagne, J.B.; Reynolds, D.B.; Yoo, J.; et al. Transcriptional regulatory code of a eukaryotic genome. Nature 2004, 431, 99–104. [Google Scholar] [CrossRef]
- Teixeira, M.C.; Monteiro, P.; Jain, P.; Tenreiro, S.; Fernandes, A.R.; Mira, N.P.; Alenquer, M.; Freitas, A.T.; Oliveira, A.L.; Sa-Correia, I. The YEASTRACT database, a tool for the analysis of transcription regulatory associations in Saccharomyces cerevisiae. Nucleic Acids Res. 2006, 34, D446–D451. [Google Scholar] [CrossRef]
- Stark, C.; Breitkreutz, B.J.; Reguly, T.; Boucher, L.; Breitkreutz, A.; Tyers, M. BioGRID, a general repository for interaction datasets. Nucleic Acids Res. 2006, 34, D535–D539. [Google Scholar] [CrossRef]
- Ashburner, M.; Ball, C.A.; Blake, J.A.; Botstein, D.; Butler, H.; Cherry, J.M.; Davis, A.P.; Dolinski, K.; Dwight, S.S.; Eppig, J.T.; et al. Gene ontology, Tool for the unification of biology. The gene ontology consortium. Nat. Genet. 2000, 25, 25–29. [Google Scholar] [CrossRef]
- Chen, H.C.; Lee, H.C.; Lin, T.Y.; Li, W.H.; Chen, B.S. Quantitative characterization of the transcriptional regulatory network in the yeast cell cycle. Bioinformatics 2004, 20, 1914–1927. [Google Scholar] [CrossRef]
- Chen, B.S.; Wang, Y.C.; Wu, W.S.; Li, W.H. A new measure of the robustness of biochemical networks. Bioinformatics 2005, 21, 2698–2705. [Google Scholar] [CrossRef]
- Chang, W.C.; Li, C.W.; Chen, B.S. Quantitative inference of dynamic regulatory pathways via microarray data. BMC Bioinf. 2005, 6, 44–62. [Google Scholar] [CrossRef]
- Lin, L.H.; Lee, H.C.; Li, W.H.; Chen, B.S. Dynamic modeling of cis-regulatory circuits and gene expression prediction via cross-gene identification. BMC Bioinf. 2005, 6, 258–274. [Google Scholar] [CrossRef]
- Chang, Y.H.; Wang, Y.C.; Chen, B.S. Identification of transcription factor cooperativity via stochastic system model. Bioinformatics 2006, 22, 2276–2282. [Google Scholar] [CrossRef]
- Wu, W.S.; Li, W.H.; Chen, B.S. Computational reconstruction of transcriptional regulatory modules of the yeast cell cycle. BMC Bioinf. 2006, 7, 421–435. [Google Scholar] [CrossRef]
- Wu, W.S.; Li, W.H.; Chen, B.S. Identifying regulatory targets of cell cycle transcription factors using gene expression and ChIP-chip data. BMC Bioinf. 2007, 8, 188–205. [Google Scholar] [CrossRef]
- Chen, B.S.; Wang, Y.C. On the attenuation and amplification of molecular noise in genetic regulatory networks. BMC Bioinf. 2006, 7, 52–65. [Google Scholar] [CrossRef]
- Chen, B.-S.; Li, C.-W. Analysing microarray data in drug discovery using systems biology. Expert Opin. Drug Discovery 2007, 2, 755–768. [Google Scholar] [CrossRef]
- Chen, B.S.; Wu, W.S.; Wang, Y.C.; Li, W.H. On the Robust Circuit Design Schemes of Biochemical Networks, Steady-State Approach. IEEE T Biomed. Circ. S 2007, 1, 91–104. [Google Scholar] [CrossRef]
- Chen, B.S.; Chen, P.W. Robust Engineered Circuit Design Principles for Stochastic Biochemical Networks With Parameter Uncertainties and Disturbances. IEEE T Biomed. Circ. S 2008, 2, 114–132. [Google Scholar] [CrossRef]
- Chen, B.S.; Lin, Y.P. A unifying mathematical framework for genetic robustness, environmental robustness, network robustness and their trade-off on phenotype robustness in biological networks Part I, Gene regulatory networks in systems and evolutionary biology. Evol. Bioinform 2013, 9, 43–68. [Google Scholar] [CrossRef]
- Chen, B.S.; Lin, Y.P. A Unifying mathematical framework for genetic robustness, environmental robustness, Network robustness and their tradeoff on phenotype robustness in biological networks Part II, Ecological networks. Evol. Bioinform 2013, 9, 69–85. [Google Scholar] [CrossRef]
- Wang, Y.C.; Huang, S.H.; Lan, C.Y.; Chen, B.S. Prediction of phenotype-associated genes via a cellular network approach, a candida albicans infection case study. PloS One 2012, 7, e35339. [Google Scholar]
- Kuo, Z.Y.; Chuang, Y.J.; Chao, C.C.; Liu, F.C.; Lan, C.Y.; Chen, B.S. Identification of infection- and defense-related genes via a dynamic host-pathogen interaction network using a candida albicans -zebrafish infection model. J. Innate Immun. 2013, 5, 137–152. [Google Scholar] [CrossRef]
- Tu, C.T.; Chen, B.S. New measurement methods of network robustness and response ability via microarray data. PloS One 2013, 8, e55230. [Google Scholar]
- Tu, C.T.; Chen, B.S. On the Increase in Network Robustness and Decrease in Network Response Ability During the Aging Process, A Systems Biology Approach via Microarray Data. IEEE/ACM Trans. Comput. Biol. Bioinf. 2013, 10, 468–480. [Google Scholar] [CrossRef]
- Csete, M.E.; Doyle, J.C. Reverse engineering of biological complexity. Science 2002, 295, 1664–1669. [Google Scholar] [CrossRef]
- Lee, K.H.; Park, J.H.; Kim, T.Y.; Kim, H.U.; Lee, S.Y. Systems metabolic engineering of Escherichia coli for L-threonine production. Mol. Syst. Biol. 2007, 3, 149. [Google Scholar]
- Park, J.H.; Lee, S.Y.; Kim, T.Y.; Kim, H.U. Application of systems biology for bioprocess development. Trends Biotechnol 2008, 26, 404–412. [Google Scholar] [CrossRef]
- Chen, B.S.; Chang, Y.T.; Wang, Y.C. Robust H infinity-stabilization design in gene networks under stochastic molecular noises, fuzzy-interpolation approach. IEEE Trans. Syst. Man Cybern B Cybern 2008, 38, 25–42. [Google Scholar] [CrossRef]
- Chu, L.H.; Chen, B.S. Construction of a cancer-perturbed protein-protein interaction network for discovery of apoptosis drug targets. BMC Syst. Biol. 2008, 2, 56. [Google Scholar] [CrossRef]
- Chen, B.S.; Chang, C.H.; Chuang, Y.J. Robust model matching control of immune systems under environmental disturbances, Dynamic game approach. J. Theor. Biol. 2008, 253, 824–837. [Google Scholar] [CrossRef]
- Chen, B.S.; Yang, S.K.; Lan, C.Y.; Chuang, Y.J. A systems biology approach to construct the gene regulatory network of systemic inflammation via microarray and databases mining. BMC Med. Genomics 2008, 1, 46. [Google Scholar] [CrossRef]
- Chen, B.S.; Chen, P.W. On the estimation of robustness and filtering ability of dynamic biochemical networks under process delays, internal parametric perturbations and external disturbances. Math. Biosci. 2009, 222, 92–108. [Google Scholar] [CrossRef]
- Wang, Y.C.; Lan, C.Y.; Hsieh, W.P.; Murillo, L.A.; Agabian, N.; Chen, B.S. Global screening of potential Candida albicans biofilm-related transcription factors via network comparison. BMC Bioinf. 2010, 11, 53–73. [Google Scholar] [CrossRef]
- Li, C.W.; Chen, B.S. Identifying Functional Mechanisms of Gene and Protein Regulatory Networks in Response to a Broader Range of Environmental Stresses. Comp. Funct. Genom. 2010, 2010, 408705. [Google Scholar]
- Gardner, T.S.; Cantor, C.R.; Collins, J.J. Construction of a genetic toggle switch in Escherichia coli. Nature 2000, 403, 339–342. [Google Scholar] [CrossRef]
- Elowitz, M.B.; Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature 2000, 403, 335–338. [Google Scholar] [CrossRef]
- Hasty, J.; McMillen, D.; Collins, J.J. Engineered gene circuits. Nature 2002, 420, 224–230. [Google Scholar] [CrossRef]
- McDaniel, R.; Weiss, R. Advances in synthetic biology, on the path from prototypes to applications. Curr. Opin. Biotechnol. 2005, 16, 476–483. [Google Scholar] [CrossRef]
- Endy, D. Foundations for engineering biology. Nature 2005, 438, 449–453. [Google Scholar] [CrossRef]
- McAdams, H.H.; Arkin, A. It’s a noisy business! Genetic regulation at the nanomolar scale. Trends Genet. 1999, 15, 65–69. [Google Scholar] [CrossRef]
- Batt, G.; Yordanov, B.; Weiss, R.; Belta, C. Robustness analysis and tuning of synthetic gene networks. Bioinformatics 2007, 23, 2415–2422. [Google Scholar] [CrossRef]
- Chen, B.S.; Chang, C.H.; Lee, H.C. Robust synthetic biology design, stochastic game theory approach. Bioinformatics 2009, 25, 1822–1830. [Google Scholar] [CrossRef]
- Chen, B.S.; Wu, C.H. A systematic design method for robust synthetic biology to satisfy design specifications. BMC Syst. Biol. 2009, 3, 66–83. [Google Scholar] [CrossRef]
- Chen, B.S.; Chen, P.W. GA-based Design Algorithms for the Robust Synthetic Genetic Oscillators with Prescribed Amplitude, Period and Phase. Gene Regul. Syst. Bio. 2010, 4, 35–52. [Google Scholar] [CrossRef]
- Chen, B.S.; Wu, C.H. Robust optimal reference-tracking design method for stochastic synthetic biology systems, T-S fuzzy approach. IEEE T. Fuzzy Syst. 2010, 18, 1144–1159. [Google Scholar] [CrossRef]
- Wu, C.H.; Zhang, W.H.; Chen, B.S. Multiobjective H-2/H-infinity synthetic gene network design based on promoter libraries. Math. Biosci. 2011, 233, 111–125. [Google Scholar] [CrossRef]
- Wu, C.H.; Lee, H.C.; Chen, B.S. Robust synthetic gene network design via library-based search method. Bioinformatics 2011, 27, 2700–2706. [Google Scholar] [CrossRef]
- Chen, B.S.; Hsu, C.Y.; Liou, J.J. Robust design of biological circuits, Evolutionary systems biology approach. J. Biomed. Biotechnol. 2011, 2011, 30423. [Google Scholar]
- Chen, B.S.; Lin, Y.P. A unifying mathematical framework for genetic robustness, Environmental robustness, Network robustness and their trade-offs on phenotype robustness in biological networks. Part III, Synthetic gene networks in synthetic biology. Evol. Bioinform 2013, 9, 87–109. [Google Scholar] [CrossRef]
- Chen, B.S.; Hsu, C.Y. Robust synchronization control scheme of a population of nonlinear stochastic synthetic genetic oscillators under intrinsic and extrinsic molecular noise via quorum sensing. BMC Systems Biology 2012, 6, 136–150. [Google Scholar] [CrossRef]
- Kim, P.J.; Lee, D.Y.; Kim, T.Y.; Lee, K.H.; Jeong, H.; Lee, S.Y.; Park, S. Metabolite essentiality elucidates robustness of Escherichia coli metabolism. Proc. Natl. Acad. Sci. USA 2007, 104, 13638–13642. [Google Scholar]
- Segre, D.; Vitkup, D.; Church, G.M. Analysis of optimality in natural and perturbed metabolic networks. Proc. Natl. Acad. Sci. USA 2002, 99, 15112–15117. [Google Scholar] [CrossRef]
- Chandran, D.; Copeland, W.B.; Sleight, S.C.; Sauro, H.M. Mathematical modeling and synthetic biology. Drug Discovery Today 2008, 5, 299–309. [Google Scholar] [CrossRef]
- Zhou, S.; Iverson, A.G.; Grayburn, W.S. Engineering a native homoethanol pathway in Escherichia coli B for ethanol production. Biotechnol. Lett. 2008, 30, 335–342. [Google Scholar]
- Shlomi, T.; Eisenberg, Y.; Sharan, R.; Ruppin, E. A genome-scale computational study of the interplay between transcriptional regulation and metabolism. Mol. Syst. Biol. 2007, 3, 101–107. [Google Scholar]
- Jamshidi, N.; Palsson, B.O. Formulating genome-scale kinetic models in the post-genome era. Mol. Syst. Biol. 2008, 4, 171–186. [Google Scholar]
- Bonneau, R.; Facciotti, M.T.; Reiss, D.J.; Schmid, A.K.; Pan, M.; Kaur, A.; Thorsson, V.; Shannon, P.; Johnson, M.H.; Bare, J.C.; et al. A predictive model for transcriptional control of physiology in a free living cell. Cell 2007, 131, 1354–1365. [Google Scholar] [CrossRef]
- Kuepfer, L.; Sauer, U.; Parrilo, P.A. Efficient classification of complete parameter regions based on semidefinite programming. BMC Bioinf. 2007, 8, 12–22. [Google Scholar] [CrossRef]
- Yao, C.W.; Hsu, B.D.; Chen, B.S. Constructing gene regulatory networks for long term photosynthetic light acclimation in Arabidopsis thaliana. BMC Bioinf. 2011, 12, 335. [Google Scholar] [CrossRef]
- Wang, Y.C.; Chen, B.S. A network-based biomarker approach for molecular investigation and diagnosis of lung cancer. BMC Med. Genomics 2011, 4, 2–16. [Google Scholar] [CrossRef]
- Wang, Y.C.; Chen, B.S. Integrated cellular network of transcription regulations and protein-protein interactions. BMC Syst. Biol. 2010, 4, 20–36. [Google Scholar] [CrossRef]
- Shiue, E.; Prather, K.L.J. Synthetic biology devices as tools for metabolic engineering. Biochem. Eng. J. 2012, 65, 82–89. [Google Scholar] [CrossRef]
- Chen, B.S.; Zhang, W.H. Stochastic H(2)/H(infinity) control with state-dependent noise. IEEE T Automat. Contr. 2004, 49, 45–57. [Google Scholar] [CrossRef]
- Zhang, W.H.; Chen, B.S.; Tseng, C.S. Robust H-infinity filtering for nonlinear stochastic systems. IEEE T Signal. Proces. 2005, 53, 589–598. [Google Scholar] [CrossRef]
- Zhang, W.H.; Chen, B.S. State feedback H(infinity) control for a class of nonlinear stochastic systems. Siam J. Control. Optim 2006, 44, 1973–1991. [Google Scholar] [CrossRef]
- Hughes, T.R.; Marton, M.J.; Jones, A.R.; Roberts, C.J.; Stoughton, R.; Armour, C.D.; Bennett, H.A.; Coffey, E.; Dai, H.Y.; He, Y.D.D.; et al. Functional discovery via a compendium of expression profiles. Cell 2000, 102, 109–126. [Google Scholar] [CrossRef]
- Johansson, R. System Modeling & Identification; Prentice-Hall, Inc.: London, UK, 1993. [Google Scholar]
- Li, C.W.; Chu, Y.H.; Chen, B.S. Construction and clarification of dynamic gene regulatory network of cancer cell cycle via microarray data. Cancer Informatics 2006, 2, 223–241. [Google Scholar]
- Boyd, S.P. Linear Matrix Inequalities in System and Control Theory; Society for Industrial and Applied Mathematics: Philadelphia, PA, USA, 1994. [Google Scholar]
- Kitano, H. Biological robustness. Nat. Rev. Genet. 2004, 5, 826–837. [Google Scholar] [CrossRef]
- Chen, B.S.; Chen, W.H.; Wu, H.L. Robust H-2/H-infinity global linearization filter design for nonlinear stochastic systems. IEEE T Circuits-I 2009, 56, 1441–1454. [Google Scholar] [CrossRef]
- Chen, B.S.; Lee, T.S.; Feng, J.H. A nonlinear H-Infinity control design in robotic systems under parameter perturbation and external disturbance. Int J. Control. 1994, 59, 439–461. [Google Scholar] [CrossRef]
- Chen, B.S.; Tseng, C.S.; Uang, H.J. Mixed H-2/H-infinity fuzzy output feedback control design for nonlinear dynamic systems, An LMI approach. IEEE T. Fuzzy Syst. 2000, 8, 249–265. [Google Scholar] [CrossRef]
- Stricker, J.; Cookson, S.; Bennett, M.R.; Mather, W.H.; Tsimring, L.S.; Hasty, J. A fast, robust and tunable synthetic gene oscillator. Nature 2008, 456, 516–519. [Google Scholar] [CrossRef]
- Tigges, M.; Marquez-Lago, T.T.; Stelling, J.; Fussenegger, M. A tunable synthetic mammalian oscillator. Nature 2009, 457, 309–312. [Google Scholar] [CrossRef]
- Basu, S.; Gerchman, Y.; Collins, C.H.; Arnold, F.H.; Weiss, R. A synthetic multicellular system for programmed pattern formation. Nature 2005, 434, 1130–1134. [Google Scholar] [CrossRef]
- Tsai, T.Y.; Choi, Y.S.; Ma, W.; Pomerening, J.R.; Tang, C.; Ferrell, J.E., Jr. Robust, Tunable biological oscillations from interlinked positive and negative feedback loops. Science 2008, 321, 126–129. [Google Scholar] [CrossRef]
- Alon, U. An Introduction to Systems Biology, Design Principles of Biological Circuits; Chapman & Hall/CRC: Boca Raton, FL, USA, 2007. [Google Scholar]
- Ellis, T.; Wang, X.; Collins, J.J. Diversity-based, Model-guided construction of synthetic gene networks with predicted functions. Nat. Biotechnol 2009, 27, 465–471. [Google Scholar] [CrossRef]
- Mondragon-Palomino, O.; Danino, T.; Selimkhanov, J.; Tsimring, L.; Hasty, J. Entrainment of a Population of Synthetic Genetic Oscillators. Science 2011, 333, 1315–1319. [Google Scholar] [CrossRef]
- Ghosh, S.; Matsuoka, Y.; Asai, Y.; Hsin, K.Y.; Kitano, H. Software for systems biology, from tools to integrated platforms. Nat. Rev. Genet. 2011, 12, 821–832. [Google Scholar]
- Tsai, K.Y.; Wang, F.S. Evolutionary optimization with data collocation for reverse engineering of biological networks. Bioinformatics 2005, 21, 1180–1188. [Google Scholar] [CrossRef]
- Chen, B.S.; Tseng, C.S.; Uang, H.J. Robustness design of nonlinear dynamic systems via fuzzy linear control. IEEE T. Fuzzy Syst. 1999, 7, 571–585. [Google Scholar] [CrossRef]
- Kitano, H. Computational systems biology. Nature 2002, 420, 206–210. [Google Scholar] [CrossRef]
© 2013 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).