Professional Documents
Culture Documents
Optimisation of Pesticide Crystal Protein Production From Bacillus Thuringiensis Employing Artificial Intelligence Techniques
Optimisation of Pesticide Crystal Protein Production From Bacillus Thuringiensis Employing Artificial Intelligence Techniques
1, 2009
77
78
Introduction
79
i, j
activation function to give its output and the output of each such neuron in the hidden
layer becomes the input of the output layer neuron after being properly weighed using the
output weights. The desired output of the ANN is therefore the weighted sum of these
output layers inputs obtained after passing through the output neurons activation
function. The logistic sigmoid function, which is the most commonly used activation
function, is given by the expression:
f ( x) =
Figure 1
1
1+e x
(1)
The network internal parameters involve: number of neurons in the hidden layer, the
activation function, the learning rate of the network, epoch size, momentum term,
tolerance and training count. The best values of these parameters, necessary for
developing good network architecture, are usually obtained by trial and error method. An
ANN model developed in this way can then be used as an objective function for
optimisation studies to find the possible input combinations that yields a maximum
response by experimentation. However, the major disadvantage with conventional
optimisation methods is that convergence to global optimum values are always based on
proximity of the initial guess values of the factors to the optimum levels, thereby
necessitating more robust optimisation techniques. In such instances, the ANN model
equation can be easily optimised using genetic algorithms, which ensures global
maximum.
80
Methodology
81
determination (R2) values during both training and validation of the network, and these
values are presented in Table 1.
Table 1
Parameters
Test
matrix
No. of hidden
layer neurons
Error
tolerance
Theta
Learning
rate
15
0.001
0.5
Values
GA parameters set for finding the optimum levels of the factors in the study
Total no. of
generations
Crossover
probability
Mutation
probability
Total
string
length
No. of binary
coded
variables
Total
no. of
runs
250
0.9
0.05
45
100
Table 3
Factors
Lower bound
Upper bound
String length
pH
3.64
10.36
15
0.98
6.02
15
1.59
18.41
15
82
Weights of the trained neural network for the prediction of PCP production by
B. thuringiensis
Wih
Neurons
pH
Who
Inoculum size
Sugar
concentration
Theta
Neurons
OD600
0.286
0.219
0.897
0.171
5.61
0.0115
0.587
2.78
2.53
7.70
7.98
6.41
10.5
0.109
9.03
0.135
7.14
3.19
0.0972
1.89
Theta
0.13
83
Residual plot of measured and predicted OD600 based on the ANN model
0.04
0.02
Residuals
0
0
10
15
20
-0.02
-0.04
-0.06
-0.08
Observations
Table 5
23 CCD showing experimental, ANN and regression predicted OD600 of the culture
OD600 values
pH
5
Inoculum size
(%)
Sugar concentration
(g/L)
Experimental
ANN
Regression
0.242
0.221
0.281
Model predicted
0.228
0.217
0.213
0.428
0.431
0.384
0.258
0.276
0.277
15
0.354
0.352
0.337
15
0.353
0.365
0.4
15
0.45
0.463
0.408
9
3.64
15
0.528
0.468
0.492
3.5
10
0.356
0.339
0.359
10.36
3.5
10
0.331
0.333
0.322
0.98
10
0.301
0.229
0.269
6.02
10
0.407
0.395
0.433
3.5
1.59
0.258
0.241
0.259
3.5
18.41
0.494
0.485
0.487
3.5
10
0.319
0.325
0.323
3.5
10
0.329
0.325
0.323
3.5
10
0.314
0.325
0.323
3.5
10
0.339
0.325
0.323
3.5
10
0.324
0.325
0.323
3.5
10
0.316
0.325
0.323
84
Figure 3
Comparison of OD600 values obtained from experiments, ANN and regression based
models
0.6
Experimental
ANN
0.5
Regression
OD 600
0.4
0.3
0.2
0.1
0
1 2 3
4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Observations
85
a predicted OD600 value of 0.582. It should be mentioned here that the RSM optimised
levels of the input factors were not in concurrence either with the experimental data or
the GA optimised levels, probably due to very low predictive accuracy (R2 < 0.85) of the
quadratic regression model, based on which RSM optimisation was performed; hence, the
result obtained using RSM may not be valid as such. This further proved the superiority
of the artificial intelligence techniques in prediction and optimisation of PCP production
using B. thuringiensis. It would also be more interesting to experimentally validate the
predicted OD600 value obtained in the present work using the ANN-GA method.
Table 6
Summary of the optimised results of OD600 of the culture using the ANN-GA method
OD600
Run 1
Run 2
Run 3
Run 4
Run 5
Maximum
0.5282
0.57643
0.52773
0.52783
0.57613
Minimum
0.35439
0.26608
0.34181
0.32586
0.25454
Average
0.50488
0.54223
0.50574
0.50536
0.53007
Figure 4
Number of
individuals
Generation no. 0
60
40
20
0
0 to 0.1
0.1 to
0.2
0.2 to
0.3
0.3 to
0.4
0.4 to
0.5
40
30
20
10
0
0.5 to
0.6
0 to 0.1
0.1 to
0.2
OD 600
Number of
individuals
Number of
Individuals
0.2 to
0.3
0.3 to
0.4
OD 600
0.4 to
0.5
0.5 to
0.6
80
60
40
20
0
0.1 to
0.2
0.3 to
0.4
OD 600
Generation no 25
0 to 0.1
0.2 to
0.3
0.4 to
0.5
0.5 to
0.6
80
60
40
20
0
0 to 0.1 0.1 to
0.2
0.2 to
0.3
0.3 to
0.4
0.4 to
0.5
0.5 to
0.6
OD 600
From the results of the hybrid approach used in this study, it could be well said that such
combinations of artificial intelligence techniques would give more insight and work
better in modelling and optimisation of other biological systems as well.
Conclusions
Earlier obtained experimental data on PCP production using Bacillus thuringiensis was
adopted for modelling and optimisation using ANN and GA respectively. The data driven
approach employing ANN for modelling the bioprocess performed excellently well with
a coefficient of determination value greater than 0.995 both during its training and
86
validation. Based on the developed ANN model, optimal values of the factors, viz., pH of
the medium, inoculum size and sugar concentration, affecting the process were obtained
using GA. The maximum PCP value obtained by this ANN-GA approach at the
optimised settings of the factors was found to be 9.17% higher compared to that obtained
from experiments.
References
Deb, K. (1995) Optimization for Engineering Design: Algorithms and Examples, Prentice Hall of
India Private Limited.
Elas, A., Ibarra-Berastegi, G., Arias, R. and Barona, A. (2006) Neural networks as a tool for
control and management of a biological reactor for treating hydrogen sulphide, Bioprocess
and Biosystems Engineering, Vol. 29, pp.129136.
Haider, M.A., Pakshirajan, K., Singh, A. and Chaudhry, S. (2008) Artificial neural network
genetic algorithm approach to optimize media constituents for enhancing lipase production by
a soil microorganism, Applied Biochemistry and Biotechnology, Vol. 144, pp.225235.
Hofte, H. and Whiteley, H.R. (1989) Insecticidal crystal proteins of Bacillus thuringiensis,
Microbiological Reviews, Vol. 53, pp.242255.
Hornik, K., Stinchcombe, M. and White, H. (1989) Multilayer feed forward networks as universal
approximators, Neural Networks, Vol. 2, pp.356359.
Maier, H.R. and Dandy, G.C. (1998) The effects of internal parameters and geometry on the
performance of back propagation neural networks: an empirical study, Environmental
Modelling and Software, Vol. 13, pp.193209.
Maier, H.R. and Dandy, G.C. (2001) Neural network based modeling of environmental variables:
a systematic approach, Mathematical and Computer Modelling, Vol. 33, pp.669682.
Poggio, T. and Girosi, F. (1990) Networks for approximation and learning, Proceedings of IEEE,
Vol. 78, No. 9, pp.14811497.
Prabagaran, S.R., Pakshirajan, K., Swaminathan, T. and Jayachandran, S. (2004) Media
optimization of Bacillus thuringiensis PBT-372 using response surface methodology,
Chemical and Biochemical Engineering Quarterly, Vol. 18, No. 2, pp.183187.
Reich, S.L., Gomez, D.R. and Dawiowski, L.E. (1999) Artificial neural network for the
identification of unknown air pollution sources, Atmospheric Environment, Vol. 33,
pp.30453052.
Rene, E.R., Maliyekkal, S.M., Philip, L. and Swaminathan, T. (2006) Back-propagation neural
network for performance prediction in trickling bed air biofilter, International Journal of
Environment and Pollution, Vol. 28, pp.382401.
Rummelhart, D.E., Hinton, G.E. and Williams, R.J. (1986) Learning representations by
back-propagation errors, Nature, Vol. 323, pp.533536.
Schnepf, E., Crickmore, N., Van Rie, J., Lereclus, D., Baum, J., Feitelson, J., Zeilger, D.R. and
Dean, D.H. (1998) Revision of the nomenclature for the Bacillus thuringiensis pesticidal
crystal proteins, Microbiology and Molecular Biology Reviews, Vol. 62, pp.775806.
Schuurmann, G. and Muller, G. (1994) Back-propagation neural networks recognition vs
prediction capability, Environmental Toxicology and Chemistry, Vol. 13, pp.743747.
Sharma, H.C., Sharma, K.K., Seetharama, N. and Ortiz, R. (2000) Prospects for using transgenic
resistance to insects in crops improvement, Electronic Journal of Biotechnology, Vol. 3,
No. 2, pp.2122.
Valdez-Castro, L., Baruch, I. and Barrera-Cortes, J. (2003) Neural networks applied to the
prediction of fed-batch fermentation kinetics of Bacillus thuringiensis, Bioprocess and
Biosystems Engineering, Vol. 25, pp.229233.
87