Food Bioactive Compounds against Diseases of the 21st CenturyView this Special Issue
Research Article | Open Access
Role of Feed Forward Neural Networks Coupled with Genetic Algorithm in Capitalizing of Intracellular Alpha-Galactosidase Production by Acinetobacter sp.
Alpha-galactosidase production in submerged fermentation by Acinetobacter sp. was optimized using feed forward neural networks and genetic algorithm (FFNN-GA). Six different parameters, pH, temperature, agitation speed, carbon source (raffinose), nitrogen source (tryptone), and K2HPO4, were chosen and used to construct 6-10-1 topology of feed forward neural network to study interactions between fermentation parameters and enzyme yield. The predicted values were further optimized by genetic algorithm (GA). The predictability of neural networks was further analysed by using mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and R2-value for training and testing data. Using hybrid neural networks and genetic algorithm, alpha-galactosidase production was improved from 7.5 U/mL to 10.2 U/mL.
Alpha-galactosidases (188.8.131.52) belong to the family of glycosyl hydrolases or glycosidases. These enzymes catalyze the hydrolysis of terminal alpha 1–6 linked galactose residues from simple and complex oligosaccharides and polysaccharides . They are widely distributed in plants, animals, and microorganisms. Alpha-galactosidases find potential applications in food, pharmacological, and chemical industries. The enzyme has been used in food industry for enhancing the nutritional quality of legumes by degrading galactooligosaccharides that cause gas or flatulence . It is also used to improve crystallization of sugar by removing raffinose from molasses in beet sugar industry  and in guar gum processing  and for enhancing bleaching of softwood along with mannanase in paper and pulp industry  and in processing of animal feed . In humans, mutations in gfA gene lead to Fabry’s disease, a rare X-linked recessive lysosomal storage disorder. Enzyme replacement therapy with α-galactosidase is considered a potential treatment for Fabry’s patients . In addition, the enzyme can also convert type “B” erythrocytes to type “O” erythrocytes  and is also used in xenotransplantation . Microbial sources for alpha-galactosidase are being explored because of ease of cultivation and fermentation conditions. However, for cost-effective production, fermentation medium plays a vital role in the commercial production of enzymes. The nutritional requirements of each microorganism are varied and are regulated by physiological, biochemical, and genetic makeup of the organism . Therefore optimization of fermentation medium is considered a crucial step for cost-effective production of the desired product. Traditional methods use one at a time method of approach that is laborious and time-consuming and it does not reflect interactions between different variables . Experiments based on statistical methods are considered to be more economical and effective than traditional methods in understanding interaction between variables and minimizing the number of experiments. Response surface methodology (RSM) is one of the widely used statistical methods for the optimization of medium parameters [12–17]. RSM-based models can predict the relationship between a limited number of input and output parameters and hence cannot be applied for highly nonlinear processes . Artificial neural networks (ANNs) and genetic algorithms (GAs) are termed as artificial intelligence that mimics the human with a given set of experimental criteria; GA enables identification of best alternative with goodness of fit by performing multiple random searches. ANN along with GA is widely used in various optimization studies, even in cases where the primary function under study is discontinuous, nondifferentiable, stochastic, or highly nonlinear [19, 20]. Currently, hybrid ANN-GA is being applied in optimizing various physical and nutritional fermentation parameters. This method has been applied to enhance the production of alkaline protease from Bacillus circulans and glutaminase from Bacillus subtilis [21, 22].
The present study focuses on enhancing the production of intracellular alpha-galactosidase from Acinetobacter sp. isolated from sugar cane waste by using hybrid artificial neural networks and genetic algorithm (ANN-GA). This is the first report of optimization of intracellular alpha-galactosidase using ANN-GA. A feed forward neural network (FFNN) together with back propagation was used for nonlinear modelling in this study to reduce the experimental error and subsequent optimization of enzyme production using genetic algorithm (GA).
2. Materials and Methods
Microorganisms were isolated from sugar cane waste. The isolate was observed to be gram-negative, short-rods, nonmotile, and nonspore forming bacteria. The isolate was positive for catalase and citrate but negative for nitrate reduction, H2S production, and oxidase and indole production. Based on morphological, biochemical, and 16s rRNA sequencing analysis, the isolate showing maximum intracellular alpha-galactosidase activity was identified as Acinetobacter sp. (Figure 1). The organism was grown at 36°C for 12 hours and maintained on agar slants at 4°C and was subcultured at 4-week interval.
2.2. Inoculum Preparation and Cell Lysis
A 24-hour-old culture 0.5% (w/v) inoculum was taken and inoculated into a 250 mL Erlenmeyer flask containing 100 mL sterile production media containing raffinose 25, tryptone 10, K2HPO4 10, MgSO47H2O 1, and FeSO47H2O 1 in g l−1 (pH 7.0). The inoculated culture media were incubated at 36°C for 24 hours in shaking incubator at an agitation speed of 170 rpm. Cells were harvested from broth by centrifugation at 10,000 g and washed with 20 mM Tris buffer (pH 7.0). The cells were suspended in the same buffer containing 0.3% (w/v) lysozyme, 0.1% (w/v) Triton X 100, and 1 mM PMSF and incubated for 1 hour at 30°C. The cells were further disrupted by sonication. Cell debris was removed by centrifugation (10,000 g, 20 minutes, 4°C). Alpha-galactosidase activity was measured in the supernatant.
2.3. Alpha-Galactosidase Activity
Alpha-galactosidase activity was measured according to Dey and Pridham  in a reaction system containing 550 μL of 20 mM Tris buffer (pH 7.2), 100 μL of supernatant (enzyme preparation), and 250 μL of 2 mM ρ-nitrophenyl-alpha-D-galactopyranoside (ρNPGal). The reaction mixture was incubated at 50°C for 10 minutes and the reaction was stopped by addition of 1 mL of 0.2 mM Na2CO3. The absorbance was read at 405 nm. One enzyme unit (U) is defined as the amount of enzyme required to produce one μmol of ρ-nitrophenol per minute under the above assay conditions.
2.4. Modelling and Optimization of Enzyme Production
2.4.1. Data Sets
In the present study, the most promising factors which influence the alpha-galactosidase production were optimized by using the neural networks and genetic algorithms. Based on the preliminary studies (data not shown), temperature, pH, agitation speed, raffinose, tryptone, and K2HPO4 concentrations were found to be the most important parameters that influence alpha-galactosidase production from the isolated bacterial strain. The list of selected variables with their minimum and maximum concentrations was given in Table 1. A central composite design with 50 experiments was employed in the present study (Table 2). The data was divided into two sets comprising 40 observations used for training the network and 10 data sets used as testing data. The training data was used to compute the network parameters. The testing data was used to ensure robustness of the network parameters.
|Data used for testing. |
2.4.2. Artificial Neural Networks
In the present study a multilayer perceptron (MLP) neural network was used. A feed forward neural network, which uses error backpropagation learning algorithm (BPNN), was constructed for modelling alpha-galactosidase production. The network consists of three layers of neurons, namely, an input layer, a hidden layer, and an output layer. All three layers are connected to the subsequent layers in the forward direction; the connections are termed as weights. The weights play a vital role in optimizing the data. Experimental conditions were chosen as inputs for the network whereas output is alpha-galactosidase activity. The number of the neurons in the hidden layer was optimized based on the trial and error method (examined from 3 to 18). All the data were normalized to −1 to +1. Scaled data are passed through the input layer and then data is propagated from input layer to hidden layer and finally to reach the targets (output layer) of the network. Every node in input and hidden layer is connected to the nodes in the subsequent layer. Each neuron in the hidden and output layer acts as a summing junction, which combines and modifies the inputs from the previous layer using the following equation: where is net input to node in hidden or output layer, is outputs of previous layer, is weights between the th node and th node, is number of neurons, and is the bias associated with node .
Sigmoid transfer function was used for the hidden layer and linear transfer function was used for the output layer to avoid error between observed and predicted values. During this process, Marquardt-Levenberg algorithm was used for training the network. Initially weight and bias values were taken randomly. However, in subsequent training steps, the weights and biases, in hidden and output layers, were adjusted in accordance with a convergence criterion to get the similarity in training and testing experimental values.
In order to evaluate the ANN output error, the coefficient of determination () was used, which describes the extent of variance in the modelled variables. The error was calculated based on difference between the experimental and predicted values. A popular measure such as mean squared error (MSE) or root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) was used to evaluate the ANN simulated data: where is number of experiments, is ANN predicted value, and is experimental value.
2.4.3. GA Optimization
Genetic algorithm was used to search in different subspace and to locate the global maximum on the objective function surface. Optimization was performed with FFNN output values of weights and bias using fitness function: WeightH is weight on connections between input and hidden nodes.WeightO is weight on connections between hidden and output nodes.
In this study, different parameters of GA optimization such as chromosome length as 36, population size as 36, crossover probability as 0.8, and mutation probability as 0.01 were taken. Optimum conditions were selected after evaluation of GA for 500 generations to achieve fine-tuned fermentation conditions in the given range of input parameters. Neural networks and genetic algorithm toolboxes of MATLAB 7.0 (The Mathworks, USA) were used in modeling studies.
3. Results and Discussion
In the present study, both physical and nutritional factors were chosen to optimize the enzyme production in shake flask. Table 2 depicts the experimental design along with experimental and predicted values of alpha-galactosidase production from Acinetobacter sp. From Table 2, it was observed that the enzyme production varied from the 3.3 to 7.5 U/mL under the various selected conditions. The observed minimum and maximum enzyme production indicate that the selected parameters have a greater influence on the alpha-galactosidase production. The data was further modelled with ANN and the conditions were optimized using the GA. The network was constructed by using the selected parameters and alpha-galactosidase production as input and output neurons. The selected six variables such as incubation temperature, pH, agitation speed, raffinose, tryptone, and K2HPO4 concentrations were chosen as input neurons in the input layer. Similarly the alpha-galactosidase production was set as output neuron in the output layer. The number of neurons in the hidden layer plays a vital role in the training time and generalization property of neural networks. Lesser number of neurons in the hidden layer would increase the training time whereas higher number of neurons in the hidden layer would cause overtraining and saturation of the network, which leads to false results. The number of neurons in a hidden layer depends on the complexity of the system being modelled. According to Sathish and Prakasham  the best approach to finding the optimal number of neurons in hidden layer is by trial and error method. In this study, the number of neurons in the hidden layer was varied from 3 to 18 and the optimal number chosen by the crossvalidation criterion with the number of epochs fixed at 1000 for all the structures studied. The neural network with 10 hidden neurons was found to have highest correlation and lowest MAPE and RMSE values. Figure 2 depicts the constructed neural network topology “6-10-1” neurons in input, hidden, and output layers.
The accuracy of the neural network based prediction can be calculated using the value based on the measured and predicted outputs in the training and test data. The calculated value was found to be 0.9994 indicating the model accuracy of the constructed ANN (Figure 3). Figure 2 depicts good correlation between the experimental values and ANN predicted values, suggesting the accuracy of the ANN predictability of the nonlinear systems.
Further, the predictability of the neural networks was analyzed based on the MSE, RMSE, MAE, and MAPE of the training and testing data. The overall MSE (), RMSE (), MAE (), and MAPE () of the training data suggested that the constructed network is suitable for the alpha-galactosidase production. This was further confirmed by testing data. The resultant data indicates a value of , , , and for MSE, RMSE, MAE, and MAPE, respectively.
3.1. Interaction Influence of Selected Variables on the Alpha-Galactosidase Production
Analysis of interactions between different selective process parameters provides information on the concentration mediated regulatory role of alpha-galactosidase production. Figure 4 shows the interactive influence of selected variables on alpha-galactosidase production. Figure 4(a) depicts the influence of temperature with tryptone concentration indicating that alpha-galactosidase production increases with temperature up to 37°C. Similarly, studies on enzyme production at different pH values indicated that the production is better at neutral pH or pH slightly above neutral pH (Figure 4(b)). Similar findings have been reported in the literature (temperature 34 to 38°C and pH 6.8 to 7.5) for mesophilic bacteria. Mixing of the components in the media has a significant role in the microbial enzyme synthesis and secretion into external environment . Figures 4(c) and 4(d) show the interaction influence of the agitation speed with tryptone and K2HPO4 concentration. From these graphs it can be concluded that, to achieve higher yields of the galactosidase, higher concentration of nutrients and higher agitation speed are needed. Figures 4(a), 4(c), and 4(f) depict the interaction influence of the tryptone with other process parameters as well as other nutrients. From these surface plots, it was observed that tryptone at 1–1.5% is favourable for the alpha-galactosidase production. The interaction influence of the carbon source (raffinose) and nitrogen source (tryptone) is depicted in Figure 4(f). It was observed that tryptone at 1% is suitable for the enzyme production. Figures 4(b) and 4(e) show the interaction influence of K2HPO4 with pH and raffinose.
3.2. GA Optimization and Validation Studies
The ANN output data was further optimized using GA. In order to obtain the best suitable conditions for alpha-galactosidase production, an objective function with weights and bias was used. Among 500 conditions generated by GA, four best suitable conditions were chosen and the validation experiments performed under these conditions (Table 3). From Table 3, it could be seen that the maximum intracellular alpha-galactosidase production was 10.2 U/mL which is 36% more than the maximum enzyme production in Table 2.
Alpha-galactosidase production titres vary for different microorganisms and are also influenced by microbial strain, enzyme localization, and physical and nutritional factors of fermentation medium. In the present study, the enzyme yield was increased from 7.5 (Table 2) to 10.2 U/mL (Table 3). Similar increase in enzyme production was reported in the case of Streptomyces griseoloalbus when optimized using RSM . List of statistical methods and activity yield using various microorganisms is presented in Table 4. Similar trend was reported by several researchers working with alkaline protease , L-glutaminase production , and rifamycin production .
Alpha-galactosidase production by Acinetobacter sp. was optimized by using feed forward ANN-GA approach, selecting six different medium parameters. A cogent correlation of 0.9994 was obtained for observed and predicted values. Interactions of raffinose and temperature with other variables are considered to be significant for maximum enzyme production. The hybrid FFNN-GA approach showed excellent predictable accuracy and can also be used for other bioprocess methods.
Conflict of Interests
The authors declare that there is no conflict of interests regarding the publication of this paper.
The authors Sirisha Edupuganti and Thadikamala Sathish thank Dr. R. S. Prakasham, Sr. Principal Scientist, Indian Institute of Chemical Technology, Hyderabad, for the technical support.
The supplementary data gives information about morphological, biochemical characteristics of the isolate identified as Acinetobacter sp. CBT01 isolated from the soil collected at sugar cane processing units. It is a fast growing organism and enters stationary phase at 5th of growth.
- P. M. Dey and J. B. Pridham, “Biochemistry of Investigation and management of dysphagia-galactosidases.,” Advances in Enzymology, vol. 36, pp. 91–130, 1972.
- V. H. Mulimani, “Enzymatic degradation of raffinose family sugars in chickpea flour,” World Journal of Microbiology and Biotechnology, vol. 13, no. 5, pp. 583–585, 1997.
- R. A. Korus and A. C. Olson, “The use of α galactosidase and invertase in hollow fiber reactors,” Biotechnology and Bioengineering, vol. 19, no. 1, pp. 1–8, 1977.
- P. V. Bulpin, M. J. Gidley, R. Jeffcoat, and D. R. Underwood, “Development of a biotechnological process for the modification of galactomannan polymers with plant α-galactosidase,” Carbohydrate Polymers, vol. 12, no. 2, pp. 155–168, 1990.
- J. H. Clarke, K. Davidson, J. E. Rixon et al., “A comparison of enzyme-aided bleaching of softwood paper pulp using combinations of xylanase, mannanase and α-galactosidase,” Applied Microbiology and Biotechnology, vol. 53, no. 6, pp. 661–667, 2000.
- B. Pan, D. Li, X. Piao, L. Zhang, and L. Guo, “Effect of dietary supplementation with α-galactosidase preparation and stachyose on growth performance, nutrient digestibility and intestinal bacterial populations of piglets,” Archives of Animal Nutrition, vol. 56, no. 5, pp. 327–337, 2002.
- B. Bengtsson, J. Johansson, C. Hollak, G. Linthorst, and U. FeldtRasmussen, “Enzyme replacement in Anderson-Fabry disease,” The Lancet, vol. 361, no. 9354, p. 352, 2003.
- J. Goldstein, G. Siviglia, R. Hurst, L. Lenny, and L. Reich, “Group B erythrocytes enzymatically converted to group O survive normally in A, B, and O individuals,” Science, vol. 215, no. 4529, pp. 168–170, 1982.
- S. Park, W. Kim, S. Choi, and Y. Kim, “Removal of alpha-gal epitopes from porcine aortic valve and pericardium using recombinant human alpha galactosidase A,” Journal of Korean Medical Science, vol. 24, no. 6, pp. 1126–1131, 2009.
- T. Sathish, G. S. Lakshmi, C. S. Rao, P. Brahmaiah, and R. S. Prakasham, “Mixture design as first step for improved glutaminase production in solid-state fermentation by isolated Bacillus sp. RSP-GLU,” Letters in Applied Microbiology, vol. 47, no. 4, pp. 256–262, 2008.
- Y. Mahalaxmi, T. Sathish, and R. S. Prakasham, “Development of balanced medium composition for improved rifamycin B production by isolated Amycolatopsis sp. RSP-3,” Letters in Applied Microbiology, vol. 49, no. 5, pp. 533–538, 2009.
- M. Hymavathi, T. Sathish, C. S. Rao, and R. S. Prakasham, “Enhancement of l-asparaginase production by isolated Bacillus circulans (MTCC 8574) using response surface methodology,” Applied Biochemistry and Biotechnology, vol. 159, no. 1, pp. 191–198, 2009.
- R. N. Vellanki, R. Potumarthi, and L. N. Mangamoori, “Constitutive expression and optimization of nutrients for streptokinase production by pichia pastoris using statistical methods,” Applied Biochemistry and Biotechnology, vol. 158, no. 1, pp. 25–40, 2009.
- K. K. Doddapaneni, R. Tatineni, R. Potumarthi, and L. N. Mangamoori, “Optimization of media constituents through response surface methodology for improved production of alkaline proteases by Serratia rubidaea,” Journal of Chemical Technology and Biotechnology, vol. 82, no. 8, pp. 721–729, 2007.
- S. Chennupati, R. Potumarthi, M. G. Rao, P. L. Manga, M. Sridevi, and A. Jetty, “Multiple responses optimization and modeling of lipase production by Rhodotorula mucilaginosa MTCC-8737 using response surface methodology,” Applied Biochemistry and Biotechnology, vol. 159, no. 2, pp. 317–329, 2009.
- R. Potumarthi, C. Subhakar, A. Pavani, and A. Jetty, “Evaluation of various parameters of calcium-alginate immobilization method for enhanced alkaline protease production by Bacillus licheniformis NCIM-2042 using statistical methods,” Bioresource Technology, vol. 99, no. 6, pp. 1776–1786, 2008.
- R. Potumarthi, J. Liu, H. Wan, and M. Danquah, “Surface immobilization of Rhizopus oryzae (ATCC 96382) for enhanced production of lipase enzyme by multiple responses optimization,” Asia-Pacific Journal of Chemical Engineering, vol. 7, no. 3, pp. S285–S295, 2012.
- B. Fang, H. Chen, X. Xie, N. Wan, and Z. Hu, “Using genetic algorithms coupling neural networks in a study of xylitol production: medium optimisation,” Process Biochemistry, vol. 38, no. 7, pp. 979–985, 2003.
- G. Montague and J. Morris, “Neural-network contributions in biotechnology,” Trends in Biotechnology, vol. 12, no. 8, pp. 312–324, 1994.
- P. Ilamathi, V. Selladurai, K. Balamurugan, and V. T. Sathyanathan, “ANN-GA approach for predictive modeling and optimization of NOx emission in a tangentially fired boiler,” Clean Technologies and Environmental Policy, vol. 15, no. 1, pp. 125–131, 2013.
- C. Subba Rao, T. Sathish, M. Mahalaxmi, G. Suvarna Laxmi, R. Sreenivas Rao, and R. S. Prakasham, “Modelling and optimization of fermentation factors for enhancement of alkaline protease production by isolated Bacillus circulans using feed-forward neural network and genetic algorithm,” Journal of Applied Microbiology, vol. 104, no. 3, pp. 889–898, 2008.
- T. Sathish and R. S. Prakasham, “Enrichment of glutaminase production by Bacillus subtilis RSP-GLU in submerged cultivation based on neural network - Genetic algorithm approach,” Journal of Chemical Technology and Biotechnology, vol. 85, no. 1, pp. 50–58, 2010.
- G. S. Anisha, R. K. Sukumaran, and P. Prema, “Evaluation of α-galactosidase biosynthesis by Streptomyces griseoloalbus in solid-state fermentation using response surface methodology,” Letters in Applied Microbiology, vol. 46, no. 3, pp. 338–343, 2008.
- Y. Mahalaxmi, T. Sathish, C. S. Rao, and R. S. Prakasham, “Corn husk as a novel substrate for the production of rifamycin B by isolated Amycolatopsis sp. RSP 3 under SSF,” Process Biochemistry, vol. 45, no. 1, pp. 47–53, 2010.
- G. S. Aniska, R. K. Sukumaran, and P. Prema, “Statistical optimization of αGalaetosidase production in submerged fermentation by Streptomyces griseoloalbus using response surface methodology,” Food Technology and Biotechnology, vol. 46, no. 2, pp. 171–177, 2008.
- C. Q. Liu, Q. H. Chen, B. Tang, H. Ruan, and G. Q. He, “Response surface methodology for optimizing the fermentation medium of alpha-galactosidase in solid-state fermentation,” Letters in Applied Microbiology, vol. 45, no. 2, pp. 206–212, 2007.
- C. Liu, H. Ruan, H. Shen et al., “Optimization of the fermentation medium for alpha-galactosidase production from Aspergillus foetidus ZU-G1 using response surface methodology,” Journal of Food Science, vol. 72, no. 4, pp. M120–M125, 2007.
Copyright © 2014 Sirisha Edupuganti et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.