2311767код для вставки
Modeling and Optimization of Microbial Hyaluronic Acid Production by Streptococcus zooepidemicus Using Radial Basis Function Neural Network Coupling Quantum-Behaved Particle Swarm Optimization Algorithm Long Liu School of Biotechnology, Jiangnan University, Wuxi 214122, China Key Laboratory of Industrial Biotechnology, Ministry of Education, Jiangnan University, Wuxi 214122, China Jun Sun and Wenbo Xu Institute of Information Technology, Jiangnan University, Wuxi 214122, China Guocheng Du School of Biotechnology, Jiangnan University, Wuxi 214122, China Key Laboratory of Industrial Biotechnology, Ministry of Education, Jiangnan University, Wuxi 214122, China Jian Chen Key Laboratory of Industrial Biotechnology, Ministry of Education, Jiangnan University, Wuxi 214122, China State Key Laboratory of Food Science and Technology, Jiangnan University, Wuxi 214122, China DOI 10.1002/btpr.278 Published online August 18, 2009 in Wiley InterScience (www.interscience.wiley.com). Hyaluronic acid (HA) is a natural biopolymer with unique physiochemical and biological properties and ?nds a wide range of applications in biomedical and cosmetic ?elds. It is important to increase HA production to meet the increasing HA market demand. This work is aimed to model and optimize the amino acids addition to enhance HA production of Streptococcus zooepidemicus with radial basis function (RBF) neural network coupling quantumbehaved particle swarm optimization (QPSO) algorithm. In the RBF-QPSO approach, RBF neural network is used as a bioprocess modeling tool and QPSO algorithm is applied to conduct the optimization with the established RBF neural network black model as the objective function. The predicted maximum HA yield was 6.92 g/L under the following conditions: arginine 0.062 g/L, cysteine 0.036 g/L, and lysine 0.043 g/L. The optimal amino acids addition allowed HA yield increased from 5.0 g/L of the control to 6.7 g/L in the validation experiments. Moreover, the modeling and optimization capacity of the RBF-QPSO approach was compared with that of response surface methodology (RSM). It was indicated that the RBF-QPSO approach gave a slightly better modeling and optimization result compared with RSM. The developed RBF-QPSO approach in this work may be helpful for the modeling and C 2009 optimization of the other multivariable, nonlinear, time-variant bioprocesses. V American Institute of Chemical Engineers Biotechnol. Prog., 25: 1819?1825, 2009 Keywords: hyaluronic acid, Streptococcus zooepidemicus, modeling, optimization Introduction Hyaluronic acid (HA) is a naturally occurring polymer comprising unbranched, polyanionic disaccharide units of D-glucuronic acid and N-acetyl glucosamine linked by b-(1!3) and b-(1!4) glycosidic bonds.1 With its unique physicochemical and biological properties such as viscoelastivity, high water-holding capacity, and biocompatibility,2 HA ?nds wide applications in food, biomedical, cosmetic, and healthcare ?elds.3 Conventionally, HA was extracted from animal tissues like rooster combs and now is increas- Correspondence concerning this article should be addressed to G. Du at [email protected] or J. Chen at [email protected] C 2009 American Institute of Chemical Engineers V ingly produced via microbial cultivation with high puri?cation ef?ciency and low production cost.3 HA from microbial cultivation will become the predominant source in the biomedical and cosmetic markets, and thus it is important to improve HA yield or productivity. Currently, most studies focus on the selection of high yield HA-producing strain and culture conditions optimization. Kim et al.4 obtained a high molecular mass HA-producing strain by nitrosoguanidine treatment and further investigated the in?uence of culture conditions (pH, temperature, agitation speed, aeration rate, and impeller type) on the microbial HA production. Zhang et al.5 enhanced microbial HA production by expressing polyhydroxybutyrate synthesis genes in Streptococcus zooepidemicus, and Krahulec and Krahulcova?6 improved HA yield via the selection of a b-glucuronidase de?cient strain. 1819 1820 Biotechnol. Prog., 2009, Vol. 25, No. 6 Currently, the commonly used strain for microbial HA production is Streptococcus zooepidemicus, which cannot synthesize some necessary nutrients such as amino acids,7 and thus it is necessary to supply balanced amino acids to improve HA yield and productivity. To achieve a best performance of bioprocesses, various optimization strategies were developed. The most frequently used approach is ??one at a time?? strategy.8,9 This approach is not only time consuming but also ignores the combined effect of all the factors involved especially in the complicated bioprocesses.10 Some mathematical and statistical methods such as orthogonal array design and response surface methodology (RSM) have been developed for bioprocess modeling and optimization.11 However, the level of orthogonal array design is limited and the model of RSM is only a second-order polynomial, which has a limited simulating ability to the nonlinear bioprocesses.12,13 Radial basis function (RBF) neural network has emerged as an attractive tool for developing nonlinear empirical models, especially in situations wherein the development of conventional empirical models becomes impractical.14 The ability to approximate functions to any desired degree of accuracy makes RBF neural network preferable for use as empirical models to other empirical models.15,16 Inspired by the quantum mechanics, we developed a novel variant form of particle swarm optimization (PSO) algorithm: quantum-behaved particle swarm optimization (QPSO) algorithm.17?19 QPSO algorithm is a powerful stochastic search and global optimization technique and has received a considerable attention. Compared with PSO algorithm, QPSO algorithm is globally convergent and can ?nd the global optima more ef?ciently and quickly especially for complex problems.20 In the previous work, we combined RBF neural network with QPSO algorithm (RBF-QPSO) as a novel approach for mixing performance modeling and optimization.21 In RBF-QPSO approach, RBF neural network was used as a bioprocess modeling tool and QPSO algorithm was applied to conduct the optimization with the established RBF black model as the objective function. In this work, the metabolic kinetics of amino acids during microbial HA production of Streptococcus zooepidemicus was characterized and the key amino acids for cell growth and HA synthesis were determined. Then, the RBF-QPSO approach was applied to model and optimize the key amino acids addition to improve HA productivity. The results obtained here may be helpful to improve microbial HA production on an industrial scale. RSM RSM combines statistical experimental designs and empirical regression models with a purpose of process optimization. An empirical model (usually a second-order quadratic polynomial) is used to correlate the response of the process with some independent variables. The general form of the second-order quadratic polynomial is X b i xi ώ X bi x2i ώ RBF neural network RBF neural network is structured by embedding RBF with a two-layer feed-forward neural network. The architecture of RBF neural network is shown in Figure 1. Mathematically, RBF neural network can be formulated as m X gπxή Ό kk uk πjjx ck jjή (2) kΌ1 where m is the neuron number of hidden layer, which is equal to cluster number of training set. kxckk represents the distance between the data point x and the RBF center ck. kk is the weight related with RBF center ck. Therefore, the output of RBF neural networks is a weighted sum of the hidden layer?s activation functions. Here, we adopt the most commonly used Gaussian RB functions as basis functions shown in Eq. 3: Rk πxή uk πxή Ό P m Ri πxή iΌ1 jjx ck jj2 Rk πxή Ό exp 2r2k (3) ! (4) In Equ. (4), rk indicates the width of the kth Guassian RB functions. One of the rk selection methods is shown as follows: r2k Ό 1 X jjx ck jj2 Mk x2h (5) k where hk is the kth cluster of training set, and Mk is the number of sample data in the kth cluster. The simulating accuracy of RBF neural network was evaluated by the root-mean square error (d) in Eq. 6: Model Theory y Ό b0 ώ Figure 1. Structure of RBF neural network for process modeling. X bij xi xj ώ e (1) where y is the predicted response, xi and xj stand for the independent variables, b0 is the intercept, bi and bj are regression coef?cients, and e is a random error component. dΌ v?????????????????????????????? uP un u πYi;e Yi;p ή2 tiΌ1 n (6) where Yi,e is the experimental data and Yi,p is the corresponding predicted data by the RBF model, and n is the number of the experimental data. QPSO algorithm PSO was originally proposed by Eberhart and Kennedy.22 Compared with the other evolutionary algorithms like Biotechnol. Prog., 2009, Vol. 25, No. 6 1821 ticles, run the algorithm for 50 iterations and decrease the value of contraction-expansion coef?cient a of QPSO from 1.0 to 0.5 linearly on the course of the run. Materials and Methods Microorganism and media Figure 2. Schematic representation of QPSO algorithm for process optimization. genetic algorithm, PSO algorithm has some attractive properties such as the easy implementation, high computational ef?ciency, and constructive cooperation between individuals. However, the global convergence cannot be guaranteed when executing the optimization task by PSO algorithm. In the previous work,17?19 we developed a novel variant form of PSO, QPSO algorithm. For QPSO algorithm, the particle moves according to the following equation: Xid Ό pid ajCd Xid jlnπ1=uή; u Ό Rand πή (7) where Rand () is the uniform random function within [0, 1], u is another uniformly distributed random number within [0, 1], Pid is the best previous position, and a is a parameter called contraction-expansion coef?cient, C is the mean of personal best positions among the particles, that is M 1X Pi M iΌ1 M 1X Ό Pi1 ; M iΌ1 CΌ M 1X Pi2 ; M iΌ1 :::; M 1X Pid M iΌ1 ! (8) and pid is determined by pid Ό u Pid ώ (1u)Pgd with u as a random number distributed uniformly within [0, 1]. The main advantage of QPSO over PSO is that QPSO algorithm is a global convergent algorithm, while PSO is not as demonstrated by Bergh.23 Figure 2 shows the application procedure of QPSO algorithm, that is, Step 1: Initialize the population by randomly generating the position vector Xi (i Ό 1, 2, and 3) of each particle and set Pi Ό Xi. Step 2: Evaluate the objective function value of each particle. That is, employ the particle?s position vector as the input of RBF estimator and obtain the output as the objective function value. Step 3: Update the personal best (pbest) position of each particle if the objective function value of the current position Xi is better than that of the pbest position. Step 4: Select the global best particle among the all particles. Step 5: Calculate the mean best position C according to Eq. 8. Step 6: For each dimension of each particle?s current position, execute the position update by Eq. 7. Step 7: If the stop criterion is not met, return to Step 2; or else, stop the optimization procedure and output the optimal solution (the current global best position) and its corresponding objective function value. In this work, we use 30 par- S. zooepidemicus WSH-24 used in this study was isolated by our laboratory. Fresh slants were cultured at 37 C for 12 h and were used for inoculation. Slant culture medium consisted of brain heart infusion (Difco, Detroit, MI 482327058) 37 g/L, glucose 1 g/L, yeast extract (Angel Yeast Co., Ltd, Hubei, China) 10 g/L, and agar powder 20 g/L. Seed culture medium consisted of sucrose 20 g/L, yeast extract 20 g/L, MgSO47H2O 2.0 g/L, MnSO44H2O 0.1 g/L, KH2PO4 2.0 g/L, CaCO3 20 g/L, and 1 mL trace elements solution. The trace element solution consisted of CaCl2 2.0 g/L, ZnCl2 0.046 g/L, and CuSO45H2O 0.019 g/L. Fermentation medium contained yeast extract 25 g/L, sucrose 70 g/L, K2SO4 1.3 g/L, MgSO47H2O 2.0 g/L, Na2HPO412H2O 6.2 g/L, FeSO47H2O 0.005 g/L, and 2.5 mL trace element solution (pH 7.2). Culture medium was sterilized at 121 C for 15 min. Microbial HA production by batch culture of S. zooepidemicus Microbial HA production by batch culture of S. zooepidemicus was conducted with an initial sucrose concentration of 70 g/L. One loop of cells from a fresh slant was transferred to 50-mL seed culture medium and cultured on a rotary shaker at 200 rpm and 37 C for 12 h. The seed culture was inoculated into a 7-L fermentor (Model KL-7L, K3T Ko Bio Tech, Korea) with a working volume of 4.0 L. The pH was automatically controlled at 7.0 by adding 5 mol/L NaOH solution. The agitation speed and aeration rate were 200 rpm and 0.5 vvm, respectively. Analytical methods Cell concentration was measured from the optical density of culture broth at 660 nm with a 722S spectrophotometer. HA concentration was measured by the carbazole method based on uronic acid determination.24 The free amino acids in the culture broth were analyzed according to the procedure as described previously.25 Results and Discussion Metabolic kinetics of amino acids in batch culture of S. zooepidemicus Figure 3 shows the time pro?les of microbial HA production by batch culture of S. zooepidemicus. Cells entered the exponential phase at 2 h and reached a maximum concentration of 13.2 g/L at 14 h. Sucrose concentration decreased from 70 g/L at 0 h to 2.6 g/L at 20 h. HA concentration increased from 0.30 g/L at 0 h to 5.0 g/L at 16 h. Lactic acid was the main product of batch culture of S. zooepidemicus and increased from 2.0 g/L at 0 h to 55 g/L at 20 h. HA was produced by S. zooepidemicus at a growth-associated manner at exponential phase and at a nongrowth-associated manner at stationary phase. The elongation of the exponential phase would be favorable for enhanced HA production. The maintenance time of exponential phase may be related to the metabolism of some key nutrition factors like amino 1822 Biotechnol. Prog., 2009, Vol. 25, No. 6 Figure 3. Time pro?les of cell growth (a), sucrose consumption (b), HA synthesis (c), and lactic acid production (d) during batch culture of S. zooepidemicus. Table 1. Metabolic Kinetics of Amino Acids During Batch Culture of S. zooepidemicus Culture Time (h) Amino acid Glutamic acid Histidine Arginine Tyrosine Cystine Valine Methionine Phenylalanine Leucine Lysine Aspartic acid Serine Glycine Threonine Alanine Iioleucine Proline Tryptotine 0 2 4 6 8 10 12 14 16 18 0.93 0.16 0.02 0.20 0.12 0.37 0.21 0.46 0.78 0.06 0.37 0.38 0.30 0.38 0.82 0.43 0.21 0.12 0.81 0.10 0.017 0.16 0.10 0.29 0.17 0.41 0.71 0.04 0.32 0.32 0.25 0.33 0.78 0.39 0.17 0.09 1.03 0.12 0.01 0.21 0.05 0.35 0.22 0.47 0.76 0.02 0.35 0.37 0.28 0.38 0.83 0.45 0.22 0.13 0.86 0.10 0.003 0.17 0.02 0.28 0.13 0.39 0.70 0.01 0.31 0.31 0.23 0.32 0.77 0.39 0.16 0.08 1.05 0.14 0.001 0.20 0.007 0.36 0.20 0.45 0.75 0.001 0.36 0.39 0.29 0.39 0.85 0.43 0.23 0.15 0.92 0.12 0 0.18 0.001 0.27 0.15 0.36 0.69 0 0.33 0.32 0.25 0.33 0.79 0.37 0.17 0.09 1.07 0.15 0 0.22 0 0.35 0.21 0.45 0.73 0 0.37 0.38 0.31 0.38 0.86 0.42 0.25 0.16 0.85 0.13 0 0.19 0 0.26 0.16 0.38 0.70 0 0.32 0.32 0.26 0.34 0.80 0.38 0.18 0.09 0.98 0.16 0 0.21 0 0.39 0.22 0.43 0.73 0 0.36 0.37 0.30 0.39 0.85 0.43 0.23 0.15 0.82 0.12 0 0.18 0 0.25 0.17 0.39 0.71 0 0.30 0.31 0.24 0.32 0.81 0.37 0.16 0.13 Note: the unit of the amino acid concentration is g/L. acids, which cannot be synthesized by S. zooepidemicus. Although the amino acid requirements of S. zooepidemicus have been characterized with at least 12 (arginine, cystine, glutamine, histidine, isoleucine, leucine, lysine, methionine, phenylalanine, tryptophane, tyrosine, and valine) identi?ed as essential,7 the metabolic kinetics of these amino acids was not clear. Table 1 shows the metabolic kinetics of 18 kinds of amino acids. It was indicated that most amino acids (except arginine, cystine, and lysine) maintained a dynamic balance during the batch culture of S. zooepidemicus. Most lactic acid bacteria have developed proteolytic systems allowing them to ef?ciently transport and utilize the peptides contained in complex nitrogen sources such as yeast extract used in this study.26 The amino acids including arginine, cystine, and lysine were metabolized quickly and their concentrations decreased to zero at 10 h, at which the cells entered the stationary phase. It seemed that the end of the exponential phase followed closely the depletion of the key amino acids such as arginine, cystine, and lysine. If the exponential phase could be extended via the addition of three key amino acids, then the improvement in HA yield could be achieved as HA was produced by S. zooepidemicus as a growth-associated manner at exponential phase. Figure 4 shows the in?uences of three key amino acids (arginine, cystine, and lysine) addition on HA yield. There existed an optimal concentration for three key amino acids in terms of HA yield, and the addition of arginine (0.05 g/ L), cystine (0.03 g/L), and lysine (0.04 g/L) increased HA yield by 25%, 18%, and 21%, respectively. As expected, the addition of three key amino acids extended the exponential phase and achieved the enhancement of HA yield. However, Biotechnol. Prog., 2009, Vol. 25, No. 6 1823 Figure 4. In?uence of arginine (a), cystine (b), and lysine (c) addition on HA synthesis. Table 2. Independent Variables and Experimental Design Levels for Response Surface Levels Independent Variables Coded Symbols 1.682 1 0 1 1.682 X1 X2 X3 0.03 0.02 0.01 0.04 0.03 0.02 0.05 0.04 0.03 0.06 0.05 0.04 0.07 0.06 0.05 Arg concentration (g/L) Cys concentration (g/L) Lys concentration (g/L) Table 3. Experimental and Predicted HA Yield for Constructing RSM and RBF Model Run 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Arg (g/L) Cys (g/L) Lys (g/L) 1 0 1 1 0 1 1.682 0 0 0 1 1 1.682 1 1 0 0 0 0 0 1 0 1 1 0 1 0 0 0 0 1 1 0 1 1 0 0 0 1.682 1.682 1 0 1 1 1.682 1 0 0 0 0 1 1 0 1 1 0 0 1.682 0 0 Experimental HA Yield (g/L) Predicted Yield by RSM (g/L) Predicted Yield by RBF (g/L) 5.89 6.32 5.73 5.53 5.79 5.71 5.82 6.32 6.32 6.32 5.60 5.72 5.98 6.15 5.83 6.32 6.32 6.12 5.78 5.96 5.92 6.53 5.99 5.68 6.13 5.58 6.02 6.52 6.55 6.51 5.63 5.95 6.32 6.06 5.99 5.53 5.53 5.91 5.97 6.08 5.93 6.52 5.99 5.67 6.13 5.57 6.01 6.53 6.55 6.51 5.64 5.95 6.33 6.07 5.98 6.53 6.52 5.90 5.97 6.08 0.02 0.02 0.03 0.01 0.01 0.02 0.02 0.03 0.02 0.03 0.01 0.02 0.01 0.01 0.01 0.01 0.01 0.03 0.01 0.02 Table 4. The Optimization Results of QPSO Algorithm in Accordance with the Different Spread Parameters of RBF Neural Network Spread X1 X2 X3 Y 0.5 0.6 0.7 0.8 0.9 1.0 1.01 0.056 0.032 0.035 6.72 0.058 0.035 0.037 6.75 0.059 0.038 0.036 6.78 0.060 0.039 0.045 6.80 0.061 0.037 0.040 6.83 0.062 0.038 0.042 6.89 0.062 0.036 0.043 6.92 the interactive effects of three amino acids addition were not considered and it was necessary to further optimize the key amino acids addition to improve HA yield Modeling and optimization of amino acids addition by RBF-QPSO approach In this work, the Box-Behnken design (Tables 2 and 3) was used to obtain data for the construction of RBF model. If Box-Behnken design was not used, one can conduct 15 to 30 runs of experiments under different conditions and thus obtain 15 to 30 groups of data for RBF training. The constructed black-box model of RBF was not expressed by an explicit mathematical model, and Table 3 shows the estima- tion of the established RBF black-box model to the experiment data. The root-mean square error d of RBF model was 0.023, and the determination coef?cient R2 was 0.998, indicating that 99.8% of the variability in the response could be explained by the established RBF model. With the RBF black-box model as the objective function, QPSO algorithm was applied to execute the optimization task. Table 4 shows the optimization results of QPSO algorithm in accordance with the different spread parameters of RBF neural network. With a spread parameter of 1.01, HA yield reached a maximum predicted value of 6.92 g/L when arginine, cystine, and lysine concentration were 0.062 g/L, 0.036 g/L and 0.043 g/L, respectively. Three repeated validation experiments under the optimal conditions were 1824 Biotechnol. Prog., 2009, Vol. 25, No. 6 performed, and a maximum HA yield of 6.70 g/L was obtained, verifying the effectiveness of the proposed RBFQPSO approach for bioprocess optimization. Comparison of modeling and optimization capacities of RBF-QPSO approach with RSM model The modeling and optimization capacities of RBF-QPSO approach were compared with those of RSM. By applying multiple regression analysis on the experimental data, the second-order quadratic polynomial model was constructed as shown in Equ. (9): Y Ό 5:8329 ώ 0:0526X1 ώ 0:0926X2 ώ 0:0127X3 0:1667X12 0:2179X22 0:2213X32 ώ 0:0375X1 X2 ώ 0:0425X1 X3 0:025X2 X3 π9ή where Y was HA yield, X1 was arginine concentration, X2 was cystine concentration, and X3 was lyscine concentration. Table 3 shows the modeling results of RSM model in terms of the predicted HA yield. The root-mean square error d of RSM model was 0.125, and the determination coef?cient R2 was 0.978, indicating that 97.8 % of the variability in the response could be explained by the established RSM model. From the equations derived from the differentiation of Eq. 9, the optimal values of X1, X2, and X3 in the coded units were 0.3, 0.1, and 1.477, respectively. Accordingly, the optimal concentration of three amino acids was arginine 0.053 g/L, cystine 0.039 g/L, and lysine 0.047 g/L. The predicted maximum HA yield was 6.53 g/L under the optimal conditions obtained by RSM model. Three repeated validation experiments under optimal conditions (arginine 0.053 g/L, cystine 0.039 g/L, and lysine 0.047 g/L) were conducted, and a maximum HA yield of 6.42 g/L was obtained. The performance of RBF black-box model was compared with RSM model. The root-mean square error d between the measured and predicted output by any of RSM and RBF model was employed as a means of comparison. The d by RBF in terms of HA yield was 0.023, whereas the d by RSM in terms of HA yield was 0.125. It was indicated that both models provided similar quality predictions for the above three independent variables in terms of HA yield, but RBF model gave a slightly better ?t to the measured data compared with RSM model, possibly due to the fact that the RSM model is only a second-order polynomial, which has the limited simulating ability to the nonlinear bioprocesses.16 Conclusion This work applied RBF-QPSO approach to model and optimize the amino acids addition for enhanced HA productivity. The predicted maximum HA yield by RBF-QPSO approach was 6.92 g/L, and a maximum HA yield of 6.70 g/ L was achieved in the validation experiments, enhanced by 34% compared with the control without amino acids addition. It was indicated that RBF-QPSO approach gave a slightly better modeling and optimization result compared with RSM model. The results obtained here may be helpful for the enhancement of microbial HA production on an industrial scale. Acknowledgments This project was ?nancially supported by Program for Changjiang Scholars and Innovative Research Team in University (No.IRT0532), the National Science Fund for Distinguished Young Scholars of China (No.20625619), Program for Cultivation and Innovation of Graduate Students in Jiangsu Province (CX08B_128Z), and 973 Project (2007CB714306). Literature Cited 1. Laurent TC, Laurent UB, Fraser JR. The structure and function of hyaluronan: an overview. Immunol Cell Biol. 1996;74:1?7. 2. Kogan G, S?olte?s L, Stern R, Gemeiner P. Hyaluronic acid: a natural biopolymer with a broad range of biomedical and industrial applications. Biotechnol Lett. 2007;29:17?25. 3. Fong Chong B, Blank LM, Mclaughlin R, Nielsen LK. Microbial hyaluronic acid production. Appl Microbiol Biotechnol. 2005;66:341?351. 4. Kim JH, Yoo SJ, Oh DK, Kweon YG, Park DW, Lee CH, Gil GH. Selection of a Streptococcus equi mutant and optimization of culture conditions for the production of high molecular weight hyaluronic acid. Enzyme Microb Tech. 1996;19:440?445. 5. Zhang JY, Hao N, Chen GQ. Effect of expressing polyhydroxybutyrate synthesis genes (phbCAB) in Streptococcus zooepidemicus on production of lactic acid and hyaluronic acid. Appl Microbiol Biotechnol. 2006;71:221?227. 6. Krahulec J, Krahulcova? J. Increase in hyaluronic acid production by Streptococcus equi subsp. zooepidemicus strain de?cient in b-glucuronidase in laboratory conditions. Appl Microbiol Biotechnol. 2006;71:415?422. 7. Armstrong DC, Cooney MJ, Johns MR. Growth and amino acid requirements of hyaluronic-acid-producing Streptococcus zooepidemicus. Appl Microbiol Biotechnol. 1997;47:309?312. 8. Vlassides S, Ferrier JG, Block DE. Using historical data for bioprocess optimization: modeling wine characteristics using arti?cial neural networks and archived process information. Biotechnol Bioeng. 2001;73:55?68. 9. Dutta JR, Dutta PK, Banerjee R. Optimization of culture parameters for extracellular protease production from a newly isolated Pseudomonas sp. using response surface and arti?cial neural network models. Process Biochem. 2004;39:2193?2198. 10. Sim JH, Kamaruddin AH. Optimization of acetic acid production from synthesis gas by chemolithotrophic bacterium-Clostridium aceticum using statistical approach. Bioresour Technol. 2008;99:2724?2735. 11. Kunamneni A, Singh S. Response surface optimization of enzymatic hydrolysis of maize starch for higher glucose production. Biochem Eng J. 2005;27:179?190. 12. Ceylan H, Kubilay S, Aktas N, Sahiner N. An approach for prediction of optimum reaction conditions for laccase-catalyzed bio-transformation of 1-naphthol by response surface methodology (RSM). Bioresour Technol. 2008;99:2025?2031. 13. Chang SW, Shaw JF, Yang KH, Chang SF, Shieh CJ. Studies of optimum conditions for covalent immobilization of Candida rugosa lipase on poly(gamma-glutamic acid) by RSM. Bioresour Technol. 2008;99:2800?2805. 14. Patnaik PR. Synthesizing cellular intelligence and arti?cial intelligence for bioprocesses. Biotechnol Adv. 2006;24:129?133. 15. Basri M, Rahman RN, Ebrahimpour A, Salleh AB, Gunawan ER, Rahman MB. Comparison of estimation capabilities of response surface methodology (RSM) with arti?cial neural network (ANN) in lipase-catalyzed synthesis of palm-based wax ester. BMC Biotechnol. 2007;7:53. 16. Huang J, Mei LH, Xia J. Application of arti?cial neural network coupling particle swarm optimization algorithm to biocatalytic production of GABA. Biotechnol Bioeng. 2006;96:924?931. 17. Sun J, Feng B, Xu WB. Particle swarm optimization with particles having quantum behavior. In: Proceedings of 2004 Congress on Evolutionary Computation, Piscataway, NJ. 2004: 325?331. 18. Sun J, Xu WB, Feng B. A global search strategy of quantumbehaved particle swarm optimization. In: Proceedings of 2004 Biotechnol. Prog., 2009, Vol. 25, No. 6 19. 20. 21. 22. IEEE Conference on Cybernetics and Intelligent Systems, Singapore. 2004:111?116. Sun J, Xu WB, Feng B. Adaptive parameter control for quantum behaved particle swarm optimization on individual level. In: Proceedings of 2005 IEEE International Conference on Systems, Man and Cybernetics, Piscataway, NJ. 2005:3049?3054. Sun J, Liu J, Xu WB. Using quantum-behaved particle swarm optimization algorithm to solve non-linear programming problems. Int J Comput Math. 2007;84:261?272. Liu L, Sun J, Zhang DX, Du GC, Chen J, Xu WB. Culture conditions optimization of microbial hyaluronic acid production by Streptococcus zooepidemicus based on radial basis function neural network and quantum behaved particle swarm optimization algorithm. Enzyme Microbial Technol. 2009;44:24?32. Eberhart RC, Kennedy J. A new optimizer using particle swarm theory. In: Proceedings of 6th International Symposium on 1825 23. 24. 25. 26. Micro Machine and Human Science, Piscataway. Nagoya, Japan: IEEE Service Center; 1995:39?43. Bergh FVd. An Analysis of Particle Swarm Optimizers. PhD Thesis. Pretoria: University of Pretoria; 2001:85?90. Bitter T, Muir HM. A modi?ed uronic acid carbazole reaction. Anal Biochem. 1962;4:330?334. Fong Chong B. Improving the Cellular Economy of Streptococcus zooepidemicus Through Metabolic Engineering. PhD Thesis. Queensland: The University of Queensland; 2002:90?93. Christensen JE, Dudley EG, Pederson JA, Steele JL. Peptidases and amino acid catabolism in lactic acid bacteria. Antonie Van Leeuwenhoek. 1999;76:217?246. Manuscript received Nov. 24, 2008, and revision received May 18, 2009.