research article role of feed forward neural networks...

Research ArticleRole of Feed Forward Neural Networks Coupled with GeneticAlgorithm in Capitalizing of Intracellular Alpha-GalactosidaseProduction by Acinetobacter sp

Sirisha Edupuganti1 Ravichandra Potumarthi2

Thadikamala Sathish3 and Lakshmi Narasu Mangamoori1

1 Centre for Biotechnology Institute of Science and Technology Jawaharlal Nehru Technological University HyderabadAndhra Pradesh 500 085 India

2Department of Chemical Engineering Monash University Clayton VIC 3800 Australia3 Bioengineering and Environmental Centre Indian Institute of Chemical Technology Hyderabad Andhra Pradesh 500 607 India

Correspondence should be addressed to Lakshmi Narasu Mangamoori mangamoorigmailcom

Received 28 February 2014 Accepted 14 July 2014 Published 31 August 2014

Academic Editor Juliana Maria Leite Nobrega de Moura Bell

Copyright copy 2014 Sirisha Edupuganti et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

Alpha-galactosidase production in submerged fermentation byAcinetobacter sp was optimized using feed forward neural networksand genetic algorithm (FFNN-GA) Six different parameters pH temperature agitation speed carbon source (raffinose) nitrogensource (tryptone) and K

2HPO4 were chosen and used to construct 6-10-1 topology of feed forward neural network to study

interactions between fermentation parameters and enzyme yieldThe 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 Usinghybrid neural networks and genetic algorithm alpha-galactosidase production was improved from 75UmL to 102UmL

1 Introduction

Alpha-galactosidases (32122) belong to the family of gly-cosyl hydrolases or glycosidases These enzymes catalyze thehydrolysis of terminal alpha 1ndash6 linked galactose residuesfrom simple and complex oligosaccharides and polysaccha-rides [1] They are widely distributed in plants animals andmicroorganismsAlpha-galactosidases findpotential applica-tions in food pharmacological and chemical industries Theenzyme has been used in food industry for enhancing thenutritional quality of legumes by degrading galactooligosac-charides that cause gas or flatulence [2] It is also used toimprove crystallization of sugar by removing raffinose frommolasses in beet sugar industry [3] and in guar gum process-ing [4] and for enhancing bleaching of softwood along withmannanase in paper and pulp industry [5] and in processingof animal feed [6] In humans mutations in gfA gene leadto Fabryrsquos disease a rare X-linked recessive lysosomal storage

disorder Enzyme replacement therapy with 120572-galactosidaseis considered a potential treatment for Fabryrsquos patients [7] Inaddition the enzyme can also convert type ldquoBrdquo erythrocytesto type ldquoOrdquo erythrocytes [8] and is also used in xenotrans-plantation [9] Microbial sources for alpha-galactosidase arebeing explored because of ease of cultivation and fermen-tation conditions However for cost-effective productionfermentation medium plays a vital role in the commercialproduction of enzymesThe nutritional requirements of eachmicroorganism are varied and are regulated by physiologicalbiochemical and geneticmakeup of the organism [10]There-fore optimization of fermentation medium is considereda crucial step for cost-effective production of the desiredproduct Traditional methods use one at a time methodof approach that is laborious and time-consuming and itdoes not reflect interactions between different variables [11]Experiments based on statistical methods are considered tobemore economical and effective than traditionalmethods in

Hindawi Publishing CorporationBioMed Research InternationalVolume 2014 Article ID 361732 8 pageshttpdxdoiorg1011552014361732

2 BioMed Research International

039

2

448458734997

999997

620

559921

NA154

652500

Nucleotide substitutions (times100)Bootstrap trials = 1000 seed = 111

Acinetobacter sp 5g (EU9167111)Acinetobacter sp V4ME25 (AJ2447651)Arctic sea ice bacterium ARK10033 (AF4683861)Gamma proteobacterium HTA554 (AB0026561)Acinetobacter sp 6-12(2010) (HM4899551)Acinetobacter sp YY-5 (AY6393761)Acinetobacter haemolyticus strain NK2BH-7 (EU3527641)Acinetobacter sp TDSAS2-27 (GQ2845301)Acinetobacter schindleri strain EH74 (GU3392991)

Acinetobacter sp EH65 (GU3392901)Acinetobacter sp BD189 (HM1202591)Acinetobacter schindleri strain LUH4760 (AJ2750412)Acinetobacter sp CBT01Acinetobacter sp 4-2 (EU5945611)Acinetobacter sp TSIW21 (EU0004491)

Figure 1 Phylogenetic analysis ofAcinetobacter sp CBT01 by ClustalW software (Accelrys San Diego CA USA)The branching pattern wasgenerated by the neighbor joining method

understanding interaction between variables andminimizingthe number of experiments Response surface methodology(RSM) is one of the widely used statistical methods forthe optimization of medium parameters [12ndash17] RSM-basedmodels can predict the relationship between a limited num-ber of input and output parameters and hence cannot beapplied for highly nonlinear processes [18] Artificial neuralnetworks (ANNs) and genetic algorithms (GAs) are termedas artificial intelligence that mimics the human with a givenset of experimental criteria GA enables identification ofbest alternative with goodness of fit by performing multiplerandom searches ANN along with GA is widely used invarious optimization studies even in cases where the primaryfunction under study is discontinuous nondifferentiablestochastic or highly nonlinear [19 20] Currently hybridANN-GA is being applied in optimizing various physicaland nutritional fermentation parameters This method hasbeen applied to enhance the production of alkaline proteasefrom Bacillus circulans and glutaminase from Bacillus subtilis[21 22]

The present study focuses on enhancing the productionof intracellular alpha-galactosidase from Acinetobacter spisolated from sugar cane waste by using hybrid artificialneural networks and genetic algorithm (ANN-GA) Thisis the first report of optimization of intracellular alpha-galactosidase usingANN-GAA feed forward neural network(FFNN) together with back propagation was used for nonlin-ear modelling in this study to reduce the experimental errorand subsequent optimization of enzyme production usinggenetic algorithm (GA)

2 Materials and Methods

21 Microorganisms Microorganisms were isolated fromsugar cane waste The isolate was observed to be gram-nega-tive short-rods nonmotile and nonspore forming bacteria

The isolate was positive for catalase and citrate but negativefor nitrate reductionH

2S production and oxidase and indole

production Based on morphological biochemical and 16srRNA sequencing analysis the isolate showing maximumintracellular alpha-galactosidase activity was identified asAcinetobacter sp (Figure 1)The organism was grown at 36∘Cfor 12 hours and maintained on agar slants at 4∘C and wassubcultured at 4-week interval

22 Inoculum Preparation and Cell Lysis A 24-hour-oldculture 05 (wv) inoculum was taken and inoculated intoa 250mL Erlenmeyer flask containing 100mL sterile pro-duction media containing raffinose 25 tryptone 10 K

2HPO4

10 MgSO4sdot7H2O 1 and FeSO

4sdot7H2O 1 in g lminus1 (pH 70)

The inoculated culture media were incubated at 36∘C for 24hours in shaking incubator at an agitation speed of 170 rpmCells were harvested from broth by centrifugation at 10000 gand washed with 20mM Tris buffer (pH 70) The cellswere suspended in the same buffer containing 03 (wv)lysozyme 01 (wv) Triton X 100 and 1mM PMSF andincubated for 1 hour at 30∘CThe cells were further disruptedby sonication Cell debris was removed by centrifugation(10000 g 20 minutes 4∘C) Alpha-galactosidase activity wasmeasured in the supernatant

23 Alpha-Galactosidase Activity Alpha-galactosidase activ-ity was measured according to Dey and Pridham [1] ina reaction system containing 550 120583L of 20mM Tris buffer(pH 72) 100 120583L of supernatant (enzyme preparation) and250 120583L of 2mM 120588-nitrophenyl-alpha-D-galactopyranoside(120588NPGal) The reaction mixture was incubated at 50∘C for10 minutes and the reaction was stopped by addition of 1mLof 02mMNa

2CO3The absorbance was read at 405 nm One

enzyme unit (U) is defined as the amount of enzyme requiredto produce one 120583mol of 120588-nitrophenol per minute under theabove assay conditions

BioMed Research International 3

Table 1 Selected factors and their minimum and maximum rangechosen for intracellular alpha-galactosidase production

Parameter Low HighTemperature (∘C) 32 40pH 6 8Agitation speed (rpm) 150 190Tryptone (g100ml) 0 2Raffinose (g100ml) 15 35

24 Modelling and Optimization of Enzyme Production241 Data Sets In the present study the most promisingfactors which influence the alpha-galactosidase productionwere optimized by using the neural networks and geneticalgorithms Based on the preliminary studies (data notshown) temperature pH agitation speed raffinose tryptoneand K

2HPO4concentrations were found to be the most

important parameters that influence alpha-galactosidase pro-duction from the isolated bacterial strain The list of selectedvariables with their minimum and maximum concentrationswas given in Table 1 A central composite design with 50experiments was employed in the present study (Table 2)Thedata was divided into two sets comprising 40 observationsused for training the network and 10 data sets used as testingdata The training data was used to compute the networkparameters The testing data was used to ensure robustnessof the network parameters

242 Artificial Neural Networks In the present study a mul-tilayer perceptron (MLP) neural network was used A feedforward neural network which uses error backpropagationlearning algorithm (BPNN) was constructed for modellingalpha-galactosidase production The network consists ofthree layers of neurons namely an input layer a hidden layerand an output layer All three layers are connected to thesubsequent layers in the forward direction the connectionsare termed as weights The weights play a vital role inoptimizing the data Experimental conditions were chosen asinputs for the network whereas output is alpha-galactosidaseactivity The number of the neurons in the hidden layer wasoptimized based on the trial and error method (examinedfrom 3 to 18) All the data were normalized to minus1 to +1Scaled data are passed through the input layer and then datais propagated from input layer to hidden layer and finally toreach the targets (output layer) of the network Every nodein input and hidden layer is connected to the nodes in thesubsequent layer Each neuron in the hidden and output layeracts as a summing junction which combines andmodifies theinputs from the previous layer using the following equation

119884119894=

119899

sum

119895=1

119909119894119908119894119895+ 119887119895 (1)

where 119884119894is net input to node 119895 in hidden or output layer 119883

119894

is outputs of previous layer 119882119894119895is weights between the 119894th

node and 119895th node 119899 is number of neurons and 119887119895is the bias

associated with node 119895

Sigmoid transfer function was used for the hidden layerand linear transfer function was used for the output layer toavoid error between observed and predicted values Duringthis process Marquardt-Levenberg algorithm was used fortraining the network Initially weight and bias values weretaken randomly However in subsequent training stepsthe weights and biases in hidden and output layers wereadjusted in accordancewith a convergence criterion to get thesimilarity in training and testing experimental values

In order to evaluate the ANN output error the coefficientof determination (1198772) was used which describes the extent ofvariance in the modelled variables The error was calculatedbased on difference between the experimental and predictedvalues A popularmeasure such asmean squared error (MSE)or root mean squared error (RMSE) mean absolute error(MAE) and mean absolute percentage error (MAPE) wasused to evaluate the ANN simulated data

MSE = 1119899

119899

sum

119894=1

(119910119901minus 119910119890)

2

RMSE = radic 1119899

119899

sum

119894=1

(119910119901minus 119910119890)

2

MAE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119910119901minus 119910119890

10038161003816100381610038161003816

MAPE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119910119901minus 119910119890

10038161003816100381610038161003816

119910119890

(2)

where 119899 is number of experiments 119910119901is ANN predicted

value and 119910119890is experimental value

243 GA Optimization Genetic algorithm was used tosearch in different subspace and to locate the global max-imum on the objective function surface Optimization wasperformed with FFNN output values of weights and biasusing fitness function

119884Output

=Weight119874

times (

2

1 + 119890(minus2timesWeight119867timesInput vector+Input bias(119887119897))

minus 1)

+Hidden layer bias (119887119867)

(3)

WeightH is weight on connections between inputand hidden nodes WeightO is weight on connectionsbetween hidden and output nodes

In this study different parameters of GA optimizationsuch as chromosome length (119871chr) as 36 population size(119873pop) as 36 crossover probability (119862119901) as 08 and mutation


Table 2 Experimental design and alpha-galactosidase activity (experimental and predicted) and error values

Serialnumber

Temperature pH Agitation Tryptone Raffinose K2HPO4 Alpha-galactosidase activity(∘C) (rpm) (g100mL) (g100mL) (g100mL) Observed Predicted Error

1 34 65 160 05 2 075 280 25 0282 34 65 160 05 3 125 470 478 minus0083 34 65 160 15 2 125 510 500 009lowast4 34 65 160 15 3 075 540 526 0135 34 65 180 05 2 125 570 579 minus0096 34 65 180 05 3 075 580 575 0047 34 65 180 15 2 075 590 612 minus0228 34 65 180 15 3 125 660 669 minus0099 34 75 160 05 2 125 480 478 00110 34 75 160 05 3 075 460 454 00511 34 75 160 15 2 075 520 496 02312 34 75 160 15 3 125 580 588 minus008lowast13 34 75 180 05 2 075 360 345 01414 34 75 180 05 3 125 570 592 minus02215 34 75 180 15 2 125 650 654 minus004lowast16 34 75 180 15 3 075 530 530 017 38 65 160 05 2 125 530 526 00318 38 65 160 05 3 075 650 642 00719 38 65 160 15 2 075 580 554 02520 38 65 160 15 3 125 470 481 minus01121 38 65 180 05 2 075 540 528 01122 38 65 180 05 3 125 590 610 minus020lowast23 38 65 180 15 2 125 540 542 minus00224 38 65 180 15 3 075 560 558 00125 38 75 160 05 2 075 520 507 01226 38 75 160 05 3 125 650 624 02527 38 75 160 15 2 125 560 561 minus00128 38 75 160 15 3 075 570 557 012lowast29 38 75 180 05 2 125 480 490 minus01030 38 75 180 05 3 075 450 456 minus00631 38 75 180 15 2 075 420 408 011lowast32 38 75 180 15 3 125 330 355 minus02533 32 7 170 1 25 1 510 511 minus00134 40 7 170 1 25 1 510 520 minus01035 36 6 170 1 25 1 570 574 minus004lowast36 36 8 170 1 25 1 500 507 minus00737 36 7 150 1 25 1 560 623 minus06338 36 7 190 1 25 1 710 658 05139 36 7 170 0 25 1 580 592 minus01240 36 7 170 2 25 1 650 649 0lowast41 36 7 170 1 15 1 550 589 minus03942 36 7 170 1 35 1 700 672 02743 36 7 170 1 25 05 500 565 minus06544 36 7 170 1 25 15 710 656 05345 36 7 170 1 25 1 750 732 017lowast46 36 7 170 1 25 1 720 732 minus01247 36 7 170 1 25 1 740 732 00748 36 7 170 1 25 1 730 732 minus002lowast49 36 7 170 1 25 1 740 732 00750 36 7 170 1 25 1 740 732 007lowastData used for testing


TemperaturepH

Agitation speedTryptoneRaffinoseK2HPO4

Input layer Output layer

WO

Hidden layer

WH

120572-Galactosidase activity

Figure 2 Feed forward neural network design used for optimization of alpha-galactosidase production119882119867 is weight of connection betweeninput and hidden layer and119882119874 is weight of connection between hidden and output layer

probability (119875mut) as 001 were taken Optimum conditionswere selected after evaluation of GA for 500 generations(119873119892max = 500) to achieve fine-tuned fermentation conditionsin the given range of input parameters Neural networks andgenetic algorithm toolboxes ofMATLAB70 (TheMathworksUSA) were used in modeling studies

3 Results and Discussion

In the present study both physical and nutritional factorswere chosen to optimize the enzyme production in shakeflask Table 2 depicts the experimental design along withexperimental and predicted values of alpha-galactosidaseproduction from Acinetobacter sp From Table 2 it wasobserved that the enzyme production varied from the 33 to75UmLunder the various selected conditionsThe observedminimum and maximum enzyme production indicate thatthe selected parameters have a greater influence on the alpha-galactosidase production The data was further modelledwith ANN and the conditions were optimized using theGA The network was constructed by using the selectedparameters and alpha-galactosidase production as input andoutput neurons The selected six variables such as incubationtemperature pH agitation speed raffinose tryptone andK2HPO4concentrations were chosen as input neurons in

the input layer Similarly the alpha-galactosidase productionwas set as output neuron in the output layer The number ofneurons in the hidden layer plays a vital role in the trainingtime and generalization property of neural networks Lessernumber of neurons in the hidden layer would increase thetraining time whereas higher number of neurons in thehidden layer would cause overtraining and saturation of thenetwork which leads to false results The number of neuronsin a hidden layer depends on the complexity of the systembeing modelled According to Sathish and Prakasham [22]the best approach to finding the optimal number of neuronsin hidden layer is by trial and error method In this studythe number of neurons in the hidden layer was varied from 3to 18 and the optimal number chosen by the crossvalidationcriterion with the number of epochs fixed at 1000 for allthe structures studied The neural network with 10 hiddenneurons was found to have highest correlation and lowestMAPE and RMSE values Figure 2 depicts the constructed

2

2

3

3

4

4

5

5

6

6

7

7

8

8

Observed data

AN

N p

redi

cted

dat

a

Training dataTesting dataLinear fit

Figure 3 Correlation chart for experimental and predicted valuesof alpha-galactosidase activity

neural network topology ldquo6-10-1rdquo neurons in input hiddenand output layers

The accuracy of the neural network based prediction canbe calculated using the 1198772 value based on the measuredand predicted outputs in the training and test data Thecalculated 1198772 value was found to be 09994 indicating themodel accuracy of the constructed ANN (Figure 3) Figure 2depicts good correlation between the experimental valuesand ANN predicted values suggesting the accuracy of theANN predictability of the nonlinear systems

Further the predictability of the neural networks wasanalyzed based on the MSE RMSE MAE and MAPE of thetraining and testing dataThe overall MSE (61times10minus4) RMSE(247 times 10minus2) MAE (34 times 10minus3) and MAPE (44 times 10minus4)of the training data suggested that the constructed networkis suitable for the alpha-galactosidase production This wasfurther confirmed by testing dataThe resultant data indicatesa value of 28 times 10minus4 1673 times 10minus2 14 times 10minus3 and 18 times 10minus3for MSE RMSE MAE and MAPE respectively


2

151

05

0 3234

3638

40

Temperature (∘ C)Tryptone concentration

120572-G

alac

tosid

ase a

ctiv

ity

minus10

minus5

minus5

0

0

5

5

10 10

15

(Um

L)

(wv)

(a)

4

6

8

10

12

15

1251075

05 665

7

875

6

7

8

9

10

11

K2 HPO

4 concentrationpH

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

(wv)

1 25

(b)

4

6

8

10

2

151

05

0

Tryptone concentration (wv)

120572-G

alac

tosid

ase a

ctiv

ity

2

4

6

8

10

12

(Um

L)

150160

170180

190

Agitation speed (rpm)

(c)

353

252

15

Raffinose concentration (wv)

150160

170180

190

Agitation speed (rpm)120572

-Gal

acto

sidas

e act

ivity

2

3

3

4

4

55

66

77

8 8

9

(Um

L)

(d)

15

2

1251

075

05

K2 HPO

4 concentration (wv)

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

152

25

353

Raffinose concentration

(wv)

3

4567

89

10

11

(e)

3

4567

89

10

11

215

105

015

225

353

2

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

Raffinose concentration (wv) Tryptone concentration

(wv)

(f)

Figure 4 Effect of selected parameters interactions on alpha-galactosidase activity (a) Tryptone versus temperature (b) K2HPO4versus pH

(c) tryptone versus agitation (d) raffinose versus agitation (e) K2HPO4versus raffinose (f) raffinose versus tryptone

31 Interaction Influence of Selected Variables on the Alpha-Galactosidase Production Analysis of interactions betweendifferent selective process parameters provides informationon the concentration mediated regulatory role of alpha-galactosidase production Figure 4 shows the interactiveinfluence of selected variables on alpha-galactosidase pro-duction Figure 4(a) depicts the influence of temperaturewithtryptone concentration indicating that alpha-galactosidaseproduction increases with temperature up to 37∘C Similarlystudies on enzyme production at different pH values indi-cated that the production is better at neutral pHor pH slightlyabove neutral pH (Figure 4(b)) Similar findings have beenreported in the literature (temperature 34 to 38∘C and pH68 to 75) for mesophilic bacteria Mixing of the componentsin the media has a significant role in the microbial enzyme

synthesis and secretion into external environment [22]Figures 4(c) and 4(d) show the interaction influence of theagitation speed with tryptone and K

2HPO4concentration

From these graphs it can be concluded that to achievehigher yields of the galactosidase higher concentration ofnutrients and higher agitation speed are needed Figures4(a) 4(c) and 4(f) depict the interaction influence of thetryptone with other process parameters as well as othernutrients From these surface plots it was observed thattryptone at 1ndash15 is favourable for the alpha-galactosidaseproduction The interaction influence of the carbon source(raffinose) and nitrogen source (tryptone) is depicted inFigure 4(f) It was observed that tryptone at 1 is suitablefor the enzyme production Figures 4(b) and 4(e) show theinteraction influence of K

2HPO4with pH and raffinose


Table 3 Best possible fermentation conditions and predicted and observed yields of enzyme alpha-galactosidase

Serialnumber

Temperature pH Agitation Tryptone Raffinose K2HPO4 Alpha-galactosidase activity(∘C) (rpm) (g100mL) (g100mL) (g100mL) Predicted Observed

1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98

Table 4 List of statistical methods used to enhance alpha-galactosidase production in various microorganisms

Serial number Organism name IE Type offermentation Design Design variables Activity UmL Reference

1 Streptomycesgriseoloalbus E Submerged

RSM(Box-Behnken

Design)

pH temperature inoculumsize inoculum age

incubation period agitationspeed carbon source

yeast extract MgSO4sdot7H2OFeSO4 and salinity

50UmL [23]

2 Streptomycesgriseoloalbus E Solid-state RSM Inoculum size

moisture and galactose

117Ug ofFermented drysubstrate of

soyabean flour

[25]

3 Aspergillusfoetidus ZU-G1 E Solid-state RSM

Wheat bran soybean mealKH2PO4 MnSO4sdotH2O and

CuSO4sdot5H2O

220719U g(minus1) drymatter [26]

4 Aspergillusfoetidus ZU-G1 E Submerged RSM

Soybean meal wheat branKH2PO4 FeSO4sdot7H2O and

the medium initial pH6475UmL [27]

32 GA Optimization and Validation Studies The ANNoutput data was further optimized using GA In order toobtain the best suitable conditions for alpha-galactosidaseproduction an objective function with weights and biaswas used Among 500 conditions generated by GA fourbest suitable conditions were chosen and the validationexperiments performed under these conditions (Table 3)FromTable 3 it could be seen that themaximum intracellularalpha-galactosidase production was 102UmL which is 36more than the maximum enzyme production in Table 2

Alpha-galactosidase production titres vary for differentmicroorganisms and are also influenced by microbial strainenzyme localization and physical and nutritional factorsof fermentation medium In the present study the enzymeyield was increased from 75 (Table 2) to 102UmL (Table 3)Similar increase in enzyme production was reported in thecase of Streptomyces griseoloalbus when optimized usingRSM [23] List of statistical methods and activity yield usingvariousmicroorganisms is presented in Table 4 Similar trendwas reported by several researchers working with alkalineprotease [21] L-glutaminase production [22] and rifamycinproduction [24]

4 Conclusion

Alpha-galactosidase production by Acinetobacter sp wasoptimized by using feed forward ANN-GA approach select-ing six different medium parameters A cogent correlation

of 09994 was obtained for observed and predicted val-ues Interactions of raffinose and temperature with othervariables are considered to be significant for maximumenzyme productionThe hybrid FFNN-GA approach showedexcellent predictable accuracy and can also be used for otherbioprocess methods

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgment

The authors Sirisha Edupuganti and Thadikamala Sathishthank Dr R S Prakasham Sr Principal Scientist IndianInstitute of Chemical Technology Hyderabad for the tech-nical support

References

[1] P M Dey and J B Pridham ldquoBiochemistry of Investigationand management of dysphagia-galactosidasesrdquo Advances inEnzymology vol 36 pp 91ndash130 1972

[2] V H Mulimani ldquoEnzymatic degradation of raffinose familysugars in chickpea flourrdquo World Journal of Microbiology andBiotechnology vol 13 no 5 pp 583ndash585 1997


[3] R A Korus and A C Olson ldquoThe use of 120572 galactosidaseand invertase in hollow fiber reactorsrdquo Biotechnology andBioengineering vol 19 no 1 pp 1ndash8 1977

[4] P V Bulpin M J Gidley R Jeffcoat and D R UnderwoodldquoDevelopment of a biotechnological process for the modifica-tion of galactomannan polymers with plant 120572-galactosidaserdquoCarbohydrate Polymers vol 12 no 2 pp 155ndash168 1990

[5] J H Clarke K Davidson J E Rixon et al ldquoA comparisonof enzyme-aided bleaching of softwood paper pulp usingcombinations of xylanase mannanase and 120572-galactosidaserdquoApplied Microbiology and Biotechnology vol 53 no 6 pp 661ndash667 2000

[6] B Pan D Li X Piao L Zhang and L Guo ldquoEffect ofdietary supplementation with 120572-galactosidase preparation andstachyose on growth performance nutrient digestibility andintestinal bacterial populations of pigletsrdquo Archives of AnimalNutrition vol 56 no 5 pp 327ndash337 2002

[7] B Bengtsson J Johansson C Hollak G Linthorst and UFeldtRasmussen ldquoEnzyme replacement in Anderson-FabrydiseaserdquoThe Lancet vol 361 no 9354 p 352 2003

[8] J Goldstein G Siviglia RHurst L Lenny and L Reich ldquoGroupB erythrocytes enzymatically converted to group O survivenormally in A B and O individualsrdquo Science vol 215 no 4529pp 168ndash170 1982

[9] S Park W Kim S Choi and Y Kim ldquoRemoval of alpha-gal epitopes from porcine aortic valve and pericardium usingrecombinant human alpha galactosidase Ardquo Journal of KoreanMedical Science vol 24 no 6 pp 1126ndash1131 2009

[10] T Sathish G S Lakshmi C S Rao P Brahmaiah and RS Prakasham ldquoMixture design as first step for improvedglutaminase production in solid-state fermentation by isolatedBacillus sp RSP-GLUrdquo Letters in Applied Microbiology vol 47no 4 pp 256ndash262 2008

[11] Y Mahalaxmi T Sathish and R S Prakasham ldquoDevelopmentof balanced medium composition for improved rifamycin Bproduction by isolated Amycolatopsis sp RSP-3rdquo Letters inApplied Microbiology vol 49 no 5 pp 533ndash538 2009

[12] M Hymavathi T Sathish C S Rao and R S PrakashamldquoEnhancement of l-asparaginase production by isolatedBacilluscirculans (MTCC 8574) using response surface methodologyrdquoApplied Biochemistry and Biotechnology vol 159 no 1 pp 191ndash198 2009

[13] R N Vellanki R Potumarthi and L N Mangamoori ldquoCon-stitutive expression and optimization of nutrients for streptok-inase production by pichia pastoris using statistical methodsrdquoApplied Biochemistry and Biotechnology vol 158 no 1 pp 25ndash40 2009

[14] K K Doddapaneni R Tatineni R Potumarthi and L NMangamoori ldquoOptimization of media constituents throughresponse surface methodology for improved production ofalkaline proteases by Serratia rubidaeardquo Journal of ChemicalTechnology and Biotechnology vol 82 no 8 pp 721ndash729 2007

[15] S Chennupati R Potumarthi M G Rao P L Manga MSridevi and A Jetty ldquoMultiple responses optimization andmodeling of lipase production by Rhodotorula mucilaginosaMTCC-8737 using response surface methodologyrdquo AppliedBiochemistry and Biotechnology vol 159 no 2 pp 317ndash3292009

[16] R Potumarthi C Subhakar A Pavani and A Jetty ldquoEvalu-ation of various parameters of calcium-alginate immobiliza-tion method for enhanced alkaline protease production by

Bacillus licheniformis NCIM-2042 using statistical methodsrdquoBioresource Technology vol 99 no 6 pp 1776ndash1786 2008

[17] R Potumarthi J LiuHWan andMDanquah ldquoSurface immo-bilization of Rhizopus oryzae (ATCC 96382) for enhanced pro-duction of lipase enzyme by multiple responses optimizationrdquoAsia-Pacific Journal of Chemical Engineering vol 7 no 3 ppS285ndashS295 2012

[18] B Fang H Chen X Xie N Wan and Z Hu ldquoUsing geneticalgorithms coupling neural networks in a study of xylitolproduction medium optimisationrdquo Process Biochemistry vol38 no 7 pp 979ndash985 2003

[19] G Montague and J Morris ldquoNeural-network contributions inbiotechnologyrdquo Trends in Biotechnology vol 12 no 8 pp 312ndash324 1994

[20] P Ilamathi V Selladurai K Balamurugan and V T Sathya-nathan ldquoANN-GA approach for predictive modeling and opti-mization of NOx emission in a tangentially fired boilerrdquo CleanTechnologies and Environmental Policy vol 15 no 1 pp 125ndash1312013

[21] C Subba Rao T Sathish M Mahalaxmi G Suvarna LaxmiR Sreenivas Rao and R S Prakasham ldquoModelling and opti-mization of fermentation factors for enhancement of alkalineprotease production by isolated Bacillus circulans using feed-forward neural network and genetic algorithmrdquo Journal ofApplied Microbiology vol 104 no 3 pp 889ndash898 2008

[22] T Sathish and R S Prakasham ldquoEnrichment of glutaminaseproduction by Bacillus subtilis RSP-GLU in submerged cultiva-tion based on neural network - Genetic algorithm approachrdquoJournal of Chemical Technology and Biotechnology vol 85 no 1pp 50ndash58 2010

[23] G S Anisha R K Sukumaran and P Prema ldquoEvaluation of120572-galactosidase biosynthesis by Streptomyces griseoloalbus insolid-state fermentation using response surface methodologyrdquoLetters in AppliedMicrobiology vol 46 no 3 pp 338ndash343 2008

[24] Y Mahalaxmi T Sathish C S Rao and R S PrakashamldquoCorn husk as a novel substrate for the production of rifamycinB by isolated Amycolatopsis sp RSP 3 under SSFrdquo ProcessBiochemistry vol 45 no 1 pp 47ndash53 2010

[25] G S Aniska R K Sukumaran and P Prema ldquoStatistical opti-mization of 120572Galaetosidase production in submerged fermen-tation by Streptomyces griseoloalbus using response surfacemethodologyrdquo Food Technology and Biotechnology vol 46 no2 pp 171ndash177 2008

[26] C Q Liu Q H Chen B Tang H Ruan and G Q He ldquoRespo-nse surface methodology for optimizing the fermentationmedium of alpha-galactosidase in solid-state fermentationrdquoLetters in AppliedMicrobiology vol 45 no 2 pp 206ndash212 2007

[27] C Liu H Ruan H Shen et al ldquoOptimization of the ferme-ntation medium for alpha-galactosidase production fromAspergillus foetidus ZU-G1 using response surface methodol-ogyrdquo Journal of Food Science vol 72 no 4 pp M120ndashM1252007

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Anatomy Research International

PeptidesInternational Journal of


Hindawi Publishing Corporation httpwwwhindawicom

International Journal of

Volume 2014

Zoology


Molecular Biology International

GenomicsInternational Journal of


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


BioinformaticsAdvances in

Marine BiologyJournal of



Signal TransductionJournal of


BioMed Research International

Evolutionary BiologyInternational Journal of



Biochemistry Research International

ArchaeaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Genetics Research International


Advances in

Virolog y

Hindawi Publishing Corporationhttpwwwhindawicom

Nucleic AcidsJournal of

Volume 2014

Stem CellsInternational



Enzyme Research



Microbiology


039

2

448458734997

999997

620

559921

NA154

652500

Nucleotide substitutions (times100)Bootstrap trials = 1000 seed = 111

Acinetobacter sp 5g (EU9167111)Acinetobacter sp V4ME25 (AJ2447651)Arctic sea ice bacterium ARK10033 (AF4683861)Gamma proteobacterium HTA554 (AB0026561)Acinetobacter sp 6-12(2010) (HM4899551)Acinetobacter sp YY-5 (AY6393761)Acinetobacter haemolyticus strain NK2BH-7 (EU3527641)Acinetobacter sp TDSAS2-27 (GQ2845301)Acinetobacter schindleri strain EH74 (GU3392991)

Acinetobacter sp EH65 (GU3392901)Acinetobacter sp BD189 (HM1202591)Acinetobacter schindleri strain LUH4760 (AJ2750412)Acinetobacter sp CBT01Acinetobacter sp 4-2 (EU5945611)Acinetobacter sp TSIW21 (EU0004491)

Figure 1 Phylogenetic analysis ofAcinetobacter sp CBT01 by ClustalW software (Accelrys San Diego CA USA)The branching pattern wasgenerated by the neighbor joining method

understanding interaction between variables andminimizingthe number of experiments Response surface methodology(RSM) is one of the widely used statistical methods forthe optimization of medium parameters [12ndash17] RSM-basedmodels can predict the relationship between a limited num-ber of input and output parameters and hence cannot beapplied for highly nonlinear processes [18] Artificial neuralnetworks (ANNs) and genetic algorithms (GAs) are termedas artificial intelligence that mimics the human with a givenset of experimental criteria GA enables identification ofbest alternative with goodness of fit by performing multiplerandom searches ANN along with GA is widely used invarious optimization studies even in cases where the primaryfunction under study is discontinuous nondifferentiablestochastic or highly nonlinear [19 20] Currently hybridANN-GA is being applied in optimizing various physicaland nutritional fermentation parameters This method hasbeen applied to enhance the production of alkaline proteasefrom Bacillus circulans and glutaminase from Bacillus subtilis[21 22]

The present study focuses on enhancing the productionof intracellular alpha-galactosidase from Acinetobacter spisolated from sugar cane waste by using hybrid artificialneural networks and genetic algorithm (ANN-GA) Thisis the first report of optimization of intracellular alpha-galactosidase usingANN-GAA feed forward neural network(FFNN) together with back propagation was used for nonlin-ear modelling in this study to reduce the experimental errorand subsequent optimization of enzyme production usinggenetic algorithm (GA)

2 Materials and Methods

21 Microorganisms Microorganisms were isolated fromsugar cane waste The isolate was observed to be gram-nega-tive short-rods nonmotile and nonspore forming bacteria

The isolate was positive for catalase and citrate but negativefor nitrate reductionH

2S production and oxidase and indole

production Based on morphological biochemical and 16srRNA sequencing analysis the isolate showing maximumintracellular alpha-galactosidase activity was identified asAcinetobacter sp (Figure 1)The organism was grown at 36∘Cfor 12 hours and maintained on agar slants at 4∘C and wassubcultured at 4-week interval

22 Inoculum Preparation and Cell Lysis A 24-hour-oldculture 05 (wv) inoculum was taken and inoculated intoa 250mL Erlenmeyer flask containing 100mL sterile pro-duction media containing raffinose 25 tryptone 10 K

2HPO4

10 MgSO4sdot7H2O 1 and FeSO

4sdot7H2O 1 in g lminus1 (pH 70)

The inoculated culture media were incubated at 36∘C for 24hours in shaking incubator at an agitation speed of 170 rpmCells were harvested from broth by centrifugation at 10000 gand washed with 20mM Tris buffer (pH 70) The cellswere suspended in the same buffer containing 03 (wv)lysozyme 01 (wv) Triton X 100 and 1mM PMSF andincubated for 1 hour at 30∘CThe cells were further disruptedby sonication Cell debris was removed by centrifugation(10000 g 20 minutes 4∘C) Alpha-galactosidase activity wasmeasured in the supernatant

23 Alpha-Galactosidase Activity Alpha-galactosidase activ-ity was measured according to Dey and Pridham [1] ina reaction system containing 550 120583L of 20mM Tris buffer(pH 72) 100 120583L of supernatant (enzyme preparation) and250 120583L of 2mM 120588-nitrophenyl-alpha-D-galactopyranoside(120588NPGal) The reaction mixture was incubated at 50∘C for10 minutes and the reaction was stopped by addition of 1mLof 02mMNa

2CO3The absorbance was read at 405 nm One

enzyme unit (U) is defined as the amount of enzyme requiredto produce one 120583mol of 120588-nitrophenol per minute under theabove assay conditions








119884119894=

119899

sum

119895=1

119909119894119908119894119895+ 119887119895 (1)


119894






MSE = 1119899

119899

sum

119894=1

(119910119901minus 119910119890)

2


119899

sum

119894=1

(119910119901minus 119910119890)

2

MAE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119910119901minus 119910119890

10038161003816100381610038161003816

MAPE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119910119901minus 119910119890

10038161003816100381610038161003816

119910119890

(2)




119884Output

=Weight119874

times (

2


minus 1)


(3)





Serialnumber




TemperaturepH



WO

Hidden layer

WH







2

2

3

3

4

4

5

5

6

6

7

7

8

8

Observed data

AN

N p

redi

cted

dat

a







2

151

05

0 3234

3638

40


120572-G

alac

tosid

ase a

ctiv

ity

minus10

minus5

minus5

0

0

5

5

10 10

15

(Um

L)

(wv)

(a)

4

6

8

10

12

15

1251075

05 665

7

875

6

7

8

9

10

11

K2 HPO

4 concentrationpH

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

(wv)

1 25

(b)

4

6

8

10

2

151

05

0


120572-G

alac

tosid

ase a

ctiv

ity

2

4

6

8

10

12

(Um

L)

150160

170180

190


(c)

353

252

15


150160

170180

190


-Gal

acto

sidas

e act

ivity

2

3

3

4

4

55

66

77

8 8

9

(Um

L)

(d)

15

2

1251

075

05

K2 HPO


4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

152

25

353


(wv)

3

4567

89

10

11

(e)

3

4567

89

10

11

215

105

015

225

353

2

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)


(wv)

(f)





2HPO4concentration





Serialnumber


1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98




RSM(Box-Behnken

Design)




50UmL [23]




soyabean flour

[25]



CuSO4sdot5H2O







4 Conclusion





Acknowledgment


References





































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology








119884119894=

119899

sum

119895=1

119909119894119908119894119895+ 119887119895 (1)


119894






MSE = 1119899

119899

sum

119894=1

(119910119901minus 119910119890)

2


119899

sum

119894=1

(119910119901minus 119910119890)

2

MAE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119910119901minus 119910119890

10038161003816100381610038161003816

MAPE = 1119899

119899

sum

119894=1

10038161003816100381610038161003816119910119901minus 119910119890

10038161003816100381610038161003816

119910119890

(2)




119884Output

=Weight119874

times (

2


minus 1)


(3)





Serialnumber




TemperaturepH



WO

Hidden layer

WH







2

2

3

3

4

4

5

5

6

6

7

7

8

8

Observed data

AN

N p

redi

cted

dat

a







2

151

05

0 3234

3638

40


120572-G

alac

tosid

ase a

ctiv

ity

minus10

minus5

minus5

0

0

5

5

10 10

15

(Um

L)

(wv)

(a)

4

6

8

10

12

15

1251075

05 665

7

875

6

7

8

9

10

11

K2 HPO

4 concentrationpH

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

(wv)

1 25

(b)

4

6

8

10

2

151

05

0


120572-G

alac

tosid

ase a

ctiv

ity

2

4

6

8

10

12

(Um

L)

150160

170180

190


(c)

353

252

15


150160

170180

190


-Gal

acto

sidas

e act

ivity

2

3

3

4

4

55

66

77

8 8

9

(Um

L)

(d)

15

2

1251

075

05

K2 HPO


4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

152

25

353


(wv)

3

4567

89

10

11

(e)

3

4567

89

10

11

215

105

015

225

353

2

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)


(wv)

(f)





2HPO4concentration





Serialnumber


1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98




RSM(Box-Behnken

Design)




50UmL [23]




soyabean flour

[25]



CuSO4sdot5H2O







4 Conclusion





Acknowledgment


References





































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology



Serialnumber




TemperaturepH



WO

Hidden layer

WH







2

2

3

3

4

4

5

5

6

6

7

7

8

8

Observed data

AN

N p

redi

cted

dat

a







2

151

05

0 3234

3638

40


120572-G

alac

tosid

ase a

ctiv

ity

minus10

minus5

minus5

0

0

5

5

10 10

15

(Um

L)

(wv)

(a)

4

6

8

10

12

15

1251075

05 665

7

875

6

7

8

9

10

11

K2 HPO

4 concentrationpH

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

(wv)

1 25

(b)

4

6

8

10

2

151

05

0


120572-G

alac

tosid

ase a

ctiv

ity

2

4

6

8

10

12

(Um

L)

150160

170180

190


(c)

353

252

15


150160

170180

190


-Gal

acto

sidas

e act

ivity

2

3

3

4

4

55

66

77

8 8

9

(Um

L)

(d)

15

2

1251

075

05

K2 HPO


4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

152

25

353


(wv)

3

4567

89

10

11

(e)

3

4567

89

10

11

215

105

015

225

353

2

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)


(wv)

(f)





2HPO4concentration





Serialnumber


1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98




RSM(Box-Behnken

Design)




50UmL [23]




soyabean flour

[25]



CuSO4sdot5H2O







4 Conclusion





Acknowledgment


References





































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology


TemperaturepH



WO

Hidden layer

WH







2

2

3

3

4

4

5

5

6

6

7

7

8

8

Observed data

AN

N p

redi

cted

dat

a







2

151

05

0 3234

3638

40


120572-G

alac

tosid

ase a

ctiv

ity

minus10

minus5

minus5

0

0

5

5

10 10

15

(Um

L)

(wv)

(a)

4

6

8

10

12

15

1251075

05 665

7

875

6

7

8

9

10

11

K2 HPO

4 concentrationpH

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

(wv)

1 25

(b)

4

6

8

10

2

151

05

0


120572-G

alac

tosid

ase a

ctiv

ity

2

4

6

8

10

12

(Um

L)

150160

170180

190


(c)

353

252

15


150160

170180

190


-Gal

acto

sidas

e act

ivity

2

3

3

4

4

55

66

77

8 8

9

(Um

L)

(d)

15

2

1251

075

05

K2 HPO


4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

152

25

353


(wv)

3

4567

89

10

11

(e)

3

4567

89

10

11

215

105

015

225

353

2

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)


(wv)

(f)





2HPO4concentration





Serialnumber


1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98




RSM(Box-Behnken

Design)




50UmL [23]




soyabean flour

[25]



CuSO4sdot5H2O







4 Conclusion





Acknowledgment


References





































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology


2

151

05

0 3234

3638

40


120572-G

alac

tosid

ase a

ctiv

ity

minus10

minus5

minus5

0

0

5

5

10 10

15

(Um

L)

(wv)

(a)

4

6

8

10

12

15

1251075

05 665

7

875

6

7

8

9

10

11

K2 HPO

4 concentrationpH

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

(wv)

1 25

(b)

4

6

8

10

2

151

05

0


120572-G

alac

tosid

ase a

ctiv

ity

2

4

6

8

10

12

(Um

L)

150160

170180

190


(c)

353

252

15


150160

170180

190


-Gal

acto

sidas

e act

ivity

2

3

3

4

4

55

66

77

8 8

9

(Um

L)

(d)

15

2

1251

075

05

K2 HPO


4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)

152

25

353


(wv)

3

4567

89

10

11

(e)

3

4567

89

10

11

215

105

015

225

353

2

4

6

8

10

12

120572-G

alac

tosid

ase a

ctiv

ity

(Um

L)


(wv)

(f)





2HPO4concentration





Serialnumber


1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98




RSM(Box-Behnken

Design)




50UmL [23]




soyabean flour

[25]



CuSO4sdot5H2O







4 Conclusion





Acknowledgment


References





































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology



Serialnumber


1 351 68 180 12 25 11 95 962 36 7 180 14 28 13 10 993 37 72 183 11 24 17 105 1024 365 7 175 14 25 13 94 98




RSM(Box-Behnken

Design)




50UmL [23]




soyabean flour

[25]



CuSO4sdot5H2O







4 Conclusion





Acknowledgment


References





































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology



































Volume 2014

Zoology






















Advances in

Virolog y



Volume 2014




Enzyme Research



Microbiology

research article role of feed forward neural networks...

Documents