Professional Documents
Culture Documents
Available at www.sciencedirect.com
Article history: The present investigation focuses on the development of an effective strategy to determine
Received 8 January 2009 the optimum environmental conditions leading to the maximum rate of biodegradation of
Received in revised form 2,4-DCP by coupling response surface methodology (RSM) with a developed genetic algo-
6 May 2009 rithm (GA) thereby ensuring minimum contact time. RSM is utilized to create an efficient
Accepted 9 May 2009 analytical model for biodegradation of 2,4-DCP in terms of environmental parameters: pH,
Published online 20 May 2009 temperature, enzyme activity and time of incubation. For this purpose, a number of
degradation experiments based on statistical three-level Box Behnken design methods
Keywords: were carried out. An effective response surface (RS) model is developed by carrying out
Box Behnken design experiments designed using the Box Behnken method. The RS model thus developed is
Genetic algorithm further interfaced with the GA to optimize the degradation conditions for optimum
2,4-DCP degradation with minimum contact time. The GA increases the biodegradation conditions
Response surface methodology to >99% within a time period of 8 h within the given range of experimental conditions. The
Crossover conditions obtained from GA were verified experimentally.
Mutation ª 2009 Elsevier Ltd. All rights reserved.
1. Introduction last few decades there has been a growing interest in solving
optimization problems using evolutionary and hereditary
Incessant industrialization has resulted in the efflux of an principles; these systems maintain a potential population of
array of chemicals of anthropogenic origin which pose candidate solutions to the problem at hand. In recent years,
a serious threat to the natural flora and fauna. These the subsurface simulation model has been combined with
compounds, popularly known as xenobiotic compounds are techniques of optimization to address important problems of
generally persistent and often have carcinogenic and muta- contaminated site management (Qin et al, 2009). Modeling of
genic effects (Hendricks et al, 1985; Gesto et al, 2009). The great site remediation was useful for dynamic evaluation/predic-
strides made in biotechnology to solve environmentally tion and/or real-time control of the remediation systems.
significant issues such as biodegradation of recalcitrant Campagnolo and Akgerman (1995) proposed a model for
xenobiotic compounds coupled with advances in computa- simulating soil vapor extraction systems that was used for
tional techniques have enabled workers to go in for optimi- biodegradation of petroleum-contaminated soils while
zation using computational modeling techniques for Christodoulatos and Mohiuddin (1996) suggested generalized
maximization of the processes as well as their deployment models for prediction of pentachlorophenol adsorption by
under field conditions (Sinha and Minsker, 2007). During the natural soils.
In the present study, an RS model using Box Behnken 2.2. Genetic algorithm
experimental design for predicting values of factors affecting
biodegradation of 2,4-DCP in wastewater was employed. 2,4- These systems have some selection process based on fitness
DCP is a potent pollutant that comes as an efflux from paper of individuals and some operators called evolutionary opera-
and textile industries. It is also used as soil fumigants and for tors. GA is one of the best known of these evolutionary
the manufacture of germicides. The regression model was programs (EP).
developed using statistical response surface methodology. In general a GA has five basic components: (a) a genetic
The accuracy of the RS model is verified with the experimental representation of solutions to the problem; (b) a way to create
measurement. The developed RS model is further coupled an initial population of solutions; (c) an evaluation function
with a developed GA to find the optimum environmental rating solutions in terms of their fitness; (d) genetic operators
conditions leading to the maximum degradation of 2,4-DCP (crossover and mutation) that alter the genetic compositions
with minimum contact time. Degradation conditions are of offspring during reproduction; (e) values of parameters of
represented with degradation parameters of enzyme GAs.
concentration, initial pH, temperature and incubation time. A GA has the following structure:
These factors may affect the efficiency of enzymatic treat-
ment, and their effects may be either independent or inter- procedure GA
active. Bhattacharya and Banerjee (2008) studied the begin
interaction of different parameters in the biodegradation of t ) 0;
2,4-DCP using RSM and got a regression coefficient of ~89%. In initialize P(t);
order to improve the efficacy of an enzymatic process, evaluate P(t);
response surface models can be coupled with modern opti- while (not termination condition) do
mization techniques to attain the optimal conditions for begin
enzymatic treatment (Liu et al., 2008). The predicted optimum t)tþ1
degradation condition by GA is validated with an experi- select P(t) from P(t - 1)
mental measurement. alter P(t)
Many global optimization methods have been proposed, evaluate P(t)
such as GA (Holland, 1975; Mitchell, 1996) and swarm intelli- end
gence (SI) which differ from other traditional search tech- end
niques in that they search among a population of points and
use probabilistic rather than deterministic transition rules. where P(t) is population of individuals in generation ‘t’.
The RS model and GA developed and utilized in this study In GA a fitness function or cost function was formed
present several advantages over other methods in the litera- depending upon the optimization of the objective function
ture. The RS model is a higher order and more sophisticated (Fig. 1). The fitness function was to be formed in such a way
polynomial model with greater accuracy (Myers and Mont- that the objective was maximized as the objective of the
gomery, 2002). The GA eliminates the difficulty of user defined present investigation was to get maximum degradation. The
parameters of the existing RS models. main objective of this study was to find the optimum condi-
The present study emphasizes the optimization of the tions which give maximum degradation which is at most
degradation of 2,4-DCP with minimum contact time by
coupling RSM and GA in the given set of experimental
parameters. The optimization of the biodegradation using
Selection of Individual
evolutionary and hereditary principles of the potent xenobi- Parameters
otic compounds is rarely reported in the literature. To the best
knowledge of the investigators, optimization of the biodeg-
radation of 2,4-DCP by coupling RSM with GA has not been
reported in the literature to date. Design of Experiments by Box
Behnken technique
80
2.8. Response surface methodology
60
Response surface methodology is an empirical modeling
technique used to evaluate the relationship between a set of
40
controllable experimental factors and observed results. This
optimization process involves three major steps: (i) perform-
ing statistically designed experiments; (ii) estimating the 20
where Y is the predicted response; k0 is a constant; ka, kb, kc, kd The developed second-order RSM-based model for degrada-
are the linear coefficients; kab, kac, kad, kbc, kbd, kcd are the tion was utilized to optimize the degradation process. This
cross-coefficients; kaa, kbb, kcc, kdd are the quadratic consists of finding the combination of input variables that
coefficients. result in maximum degradation. Hence, the degradation
This response is preferred because relatively few experi- optimization can be stated as follows:
mental combinations of the variables are adequate to esti-
mate potentially complex response function. A total number Find pH, temperature, time and enzyme activity
of 27 experiments were necessarily carried out to estimate the To Maximize degradation ¼ f (pH, temperature, time and enzyme
15 coefficients for the biodegradation of 2,4-DCP. Data were activity)
analyzed using the Minitab program including ANOVA to Subjected to constraint: degradation £ 100%
find out the interaction between the variables and the Parameter ranges:
response. The quality of the fit of this model was expressed by 5.5 pH 7
the coefficient of determination (R2) in the same program. 25 temperature 40
The mathematical expression of the relationship of 2,4-DCP 7 time 10
degradation with variables A, B, C and D (pH, temperature, time 5 enzyme activity 10
Table 1 – ANOVA results for the quadratic equation of Minitab for degradation of 2,4-DCP.
Source d.f. Seq. SS Adj. SS Adj. MS F P
As seen from the response surface analysis, there exists is indispensable in genetic algorithms as it helps to push out
a non-linear relationship between the degradation and a trapped solution from a local optima to a global optima.
process parameters. Further, the combinations of process Hence the probability of reaching the global optima increases
parameters required to maximize degradation are different. with the induction of randomization (Mitchell, 1996).
Thus, it requires an efficient optimization tool to determine
the combination of process parameters which maximize 3.1.4. Crossover and mutation
degradation and hence GA is employed. Crossover is a recombination operator where gene informa-
The GAs are more likely to converge to global optimum tion is exchanged at random locations between two parent
than conventional optimization techniques, since they search chromosomes, which are randomly selected high fitness
from a population of points, and are based on probabilistic valued chromosomes from the population (Gen and Cheng,
rules. The conventional optimization techniques are ordi- 1999). The crossover fraction ( pc) is also randomly selected.
narily based on deterministic hill-climbing methods, which For instance, if crossover fraction ( pc) is 0.5 then the first 30
may find local optima. The GAs can also tolerate discontinu- higher fitness valued chromosomes out of 60 chromosomes
ities and noisy function evaluations. In each cycle of genetic are to be selected for crossover when the chromosomes are
operation, termed an evolution process, a subsequent gener- sorted according to the decreasing fitness values.
ation is created from the chromosomes in the current pop- For constrained optimization the central problem is to
ulation. This consists of manipulation of genes, where genes handle the constraints as the manipulation of chromosomes in
of parents are mixed and recombined for the production of the process of crossover and mutation often yield infeasible
offspring in the next generation. This evolution process chromosomes which may give better results but violating the
consists of selection or reproduction, crossover and mutation. constraints. Four techniques can be used for handling the
constraints (Michalewicz, 1996):
3.1.1. Genes, chromosomes and fitness function
The genes are the searching parameters for the optimization Rejecting strategy: Simply discards the infeasible
problem. In the simple GA, the genes are represented with chromosomes.
finite length of binary codes, 0 and 1. The chromosomes are Repairing strategy: Repairs the infeasible chromosomes.
the strings containing defining genes. Thus, the chromosome Modifying genetic operators strategy: Modifies the problem
for the optimization consisted of four genes corresponding to representation and genetic operators (Crossover, Mutation
four searching parameters: pH, temperature, time and and Selection).
enzyme activity. Each gene was represented by six bits of Penalty strategy: Penalizing the infeasible solutions.
binary codes and hence the chromosome constructed was of
24 bits length as shown below: Penalty strategy is the most popular and common
strategy amongst the above mentioned methods. Except
Sample chromosome ¼ [110001111001011001100100] penalty strategy, others never take infeasible solution at all
into consideration. As in the constrained optimization,
The genes of each chromosome were decoded as: infeasible solutions generally take a bigger portion than
feasible solutions and in such cases feasible solutions
X ¼ DðZmax Zmin Þ= 2nb 1 þ Zmin ; are really difficult to find out if genetic search is confined to
Z ¼ pH; temperature; time and enzyme activity feasible regions only. Penalty strategy allows the movement
of genetic search in infeasible regions too to yield more rapid
where D is the decimal equivalent of binary, Zmax the optimization and produce better results than do approaches
maximum limit for Z, Zmin the minimum limit for Z, and nb is limiting search trajectories only to feasible regions of search
the number of binary bits for genes (i.e. 6). space. Hence it provides greater adaptability to biological
systems which in general have lower consistency. This
3.1.2. Fitness function technique transforms the constrained optimization problem
The uncoded response surface found earlier was used as the into a unconstrained one, in which a penalty function is
objective function which is to be maximized. A penalty func- added to the objective function for any violation of
tion was introduced in the fitness function to discard the 100% constraints (Michalewicz, 1996).
or more degradation values. The fitness function F(x) was of In the present investigation, the penalty function P(x) is
the following structure: defined as:
1
Fitness function ¼ FðxÞ ¼ þ PðxÞ
Degradation PðxÞ ¼ maximumf0; Mðf ðxÞ 100Þg
where P(x) represents the penalty function. where x is a solution to the problem, M is a large number such
as 10,000,000, and f(x) is the degradation function.
3.1.3. Selection This depicts that, whenever the degradation value is more
In the present study, the tournament selection with replace- than 100 then a large positive quantity will be added to the
ment and selection size 4 (called tournament size) is applied fitness value and as it is a minimization problem, it is auto-
for the selection of chromosomes. This method randomly matically discarded otherwise it would take a zero value when
chooses four chromosomes and picks out the best chromo- the degradation value is less than 100, hence proving its
some for reproduction based on fitness value. Randomization feasibility.
3508 water research 43 (2009) 3503–3510
Q1 ¼ F01
1 0 1 D E F G H I 0 1 1 Q2 ¼ F01 þ F02
Progeny Chromosome1
..;
Progeny Chromosome2 A B C 0 1 1 1 0 1 J K L Q100 ¼ F01 þ F02 þ ..F0100
0.003
0.004
0.006
0.007
0.009
0.010
Mutation Probability
Mutation probability=0.008
Maximum no. of generations=100
10.5 Population size varied from 100 to 1000
10.2
9.9
9.6
100
250
400
550
700
850
1000
Population size
c 11
Maximum Generation No. v/s Fitness
Fitness(10e-3)
250
350
450
550
650
750
850
950
Fig. 5 – Study of parameters for selection of factors for genetic algorithm: (a) mutation probability versus fitness; (b)
population size versus fitness; (c) maximum generation number versus fitness.
water research 43 (2009) 3503–3510 3509
parameter values of 6.10021, 36.85669 C, 7.99644 h and adding more flexibility in the model. The model was further
9.92093 IU ml1 as pH, temperature, time and enzyme activity validated and the accuracy was found to be 94.67%. This
respectively. In the real experiment, these values of the reinforces the deployment of the methodology in field condi-
parameters gave a degradation of 98.91% which was very tions for 2,4-DCP biodegradation at a concentration at which
similar to the data obtained. Bhattacharya and Banerjee (2008) the growth of whole cell organisms is not possible.
reported a maximum degradation efficiency of ~98% at a pH 6,
temperature 40 C, time 9 h and an enzyme concentration of
8 IU ml1. Zhang et al. (2008) reported 94% removal efficiency references
of 2,4-DCP within 10 h at pH 5.5 while exploring the idea of
a promotion in the rate of degradation with the increase in
laccase concentration or an increase in temperature. Using Bhattacharya, S.S., Banerjee, R., 2008. Laccase mediated
GA, >99% degradation is possible by varying the conditions. biodegradation of 2,4-DCP using response surface
The increase in the degradation rate can be attributed to the methodology. Chemosphere 73, 81–85.
greater stability of laccase at slightly higher pH. The greater Box, G.E.P., Behnken, D.W., 1960. Some new three level designs
degradation efficiency at elevated temperature might be due for the study of quantitative variables. Technometrics 2,
455–475.
to the higher collision rate between the molecules of the
Campagnolo, J.F., Akgerman, A., 1995. Modeling of soil vapor
enzyme and 2,4-DCP. From this study, it is evident that extraction (SVE) systems – part II, biodegradation aspects of
a balance has to be struck between temperatures of maximum soil vapor extraction. Waste Management 15, 391–397.
degradation of 2,4-DCP and laccase denaturation. This high- Christodoulatos, C., Mohiuddin, M., 1996. Generalized models for
lights the efficacy of the GA to overcome the limitations of prediction of pentachlorophenol adsorption by natural soils.
user defined boundary parameters. Furthermore, while Water Environment Research 68, 370–378.
varying the concentration of 2,4-DCP ranging from 50m to Ettala, M., Koskela, Kiesila, A., 1992. Removal of chlorophenols in
a municipal sewage treatment plant using activated sludge.
1000 mg ml1 to check the feasibility of the process across
Water Research 26, 797–804.
a concentration range in addition to the concentration of Gen, M., Cheng, R., 1999. Genetic Algorithms and Engineering
1 mM (163 mg l1) at which the optimum conditions of Design, first ed. John Wiley & Sons, Inc., New York.
biodegradation were established, it was observed that the Km Gesto, M., Tintos, A., Soengas, J.L., Mı́gue, J.M., 2009. b-
value obtained from the experiments is 350 mg ml1. The levels Naphthoflavone and benzo(a)pyrene alter dopaminergic,
of 2,4-DCP in the environment range from ~150 mg l1(Valo noradrenergic, and serotonergic systems in brain and
pituitary of rainbow trout (Oncorhynchus mykiss). Ecotoxicology
et al., 1990) to 150–200 mg l1 (Etalla et al., 1992). This rein-
and Environmental Safety 72, 191–198.
forces the efficacy of laccase for the biodegradation of 2,4-DCP
Hendricks, J.D., Meyers, T.R., Shelton, D.W., Casteel, J.L., Bailey, G.S.,
in field conditions. Hence, the present investigation highlights 1985. Hepatocarcinogenicity of benzo[a]pyrene to rainbow trout
the fastest biodegradation of 2,4-DCP which can be applied at by dietary exposure and intraperitoneal injection. Journal of the
a concentration where whole cells fail to thrive (Bhattacharya National Cancer Institute 74, 839–851.
and Banerjee, 2008). Holland, J.H., 1975. Adaptation in Natural and Artificial Systems,
first ed. The University of Michigan Press, Ann Arbor.
Liu, J., Guan, X., Zhu, D., Sun, J., 2008. Optimization of the
3.2. Validation of the model
enzymatic pretreatment in oat bran protein extraction by
particle swarm optimization algorithms for response surface
For a model to be declared industrially fit, it is important to modeling. LWT – Food Science and Technology 41, 1913–1918.
validate the model. A regression analysis was carried out Michalewicz, Z., 1996. Genetic Algorithms þ Data
(Table 2) and it was found that the model developed by GA has Structure ¼ Evolution Programs, third ed. Springer Verlag,
an accuracy of 94.67% which is fairly good for deployment in New York.
field conditions. Residual analysis was further carried out Mitchell, M., 1996. An introduction to Genetic Algorithms, first ed.
The MIT Press, London.
(Table 3) for the validation of the model and it was found that
Montgomery, D.C., 2004. Design and Analysis of Experiments,
all the standardized residuals fell within the range of 1 to þ2 fourth ed. John Wiley & Sons Inc., New York.
barring one observation which was 5.45. This is a peculiar case Myers, R.H., Montgomery, D.C., 2002. Response Surface
of an outlier because the predicted value exceeded 100% Methodology: Process and Product Optimization Using
degradation which is not practically feasible. This is a fairly Designed Experiments, second ed. Wiley, New York.
good result as the acceptable range of residuals is between 3 Nelson, D.L., Cox, M., 2000. Lehninger Principles of Biochemistry,
to þ3. third ed. Macmillan Worth Publishers, New York.
Qin, X.S., Huang, G.H., He, L., 2009. Simulation and optimization
technologies for petroleum waste management and
remediation process control. Journal of Environmental
4. Conclusions Management 90, 54–76.
Sinha, E., Minsker, B.S., 2007. Multiscale island injection GAs for
In the present investigation, RSM was effectively coupled with groundwater remediation. Advances in Water Resources 30,
the GA to develop a model and optimize the experimental 1933–1942.
Valo, R.J., Haggblom, M.M., Salkinoja-Salonen, M., 1990.
conditions to effect ~99% biodegradation of 2,4-DCP within
Bioremediation of Chlorophenol containing simulated ground
a time period of 8 h. This was an improvement compared to
water by immobilized bacteria. Water Research 24, 253–258.
the regression model which showed ~98% degradation within Zhang, J., Liu, X., Xu, Z., Chen, H., Yang, Y., 2008. Degradation of
9 h. This highlights the efficacy of the methodology to over- chlorophenols catalyzed by laccase. International
come the limitations of user defined parameters thereby Biodeterioration & Biodegradation 61, 351–356.