Trends in Biotechnology
Volume 17, Issue 4, 1 April 1999, Pages 155-162
Journal home page for Trends in Biotechnology

Review
Applying neural networks as software sensors for enzyme engineering

https://doi.org/10.1016/S0167-7799(98)01299-2Get rights and content

Abstract

The on-line control of enzyme-production processes is difficult, owing to the uncertainties typical of biological systems and to the lack of suitable on-line sensors for key process variables. For example, intelligent methods to predict the end point of fermentation could be of great economic value. Computer-assisted control based on artificial-neural-network models offers a novel solution in such situations. Well-trained feedforward–backpropagation neural networks can be used as software sensors in enzyme-process control; their performance can be affected by a number of factors.

Section snippets

What is a neural network?

Neural networks are universal approximators – that is, they possess the ability to approximate any real-value continuous function to any desired degree of accuracy15. They require relatively little time to construct and do not require any a priori knowledge of the relationships between the process variables in question and can, therefore, be considered to be ‘black box’ systems. As a consequence, in typical situations, they are unable to explain their actions satisfactorily, which has been said

History in a nutshell

In creating an appropriate neural-network model for a given application, a lot of intuition and trial and error are required. Furthermore, neural-network software and its running environment must be able to preprocess the data intended as an input for a certain neural-network model, to postprocess the data emerging as an output and to allow interactive data manipulation when a number of possible solutions to a problem need to be explored.

Although the principles of neurocontrol have been known

Glucoamylase

Glucoamylase is one of the key enzymes in starch processing. Using an industrial glucoamylase-producing Aspergillus niger strain grown on a starch-based medium31, a well-trained neural network of 4–10–6 topology performed satisfactorily, especially in the prediction of glucoamylase activity. High coefficients of determination were obtained (R2 = 0.994 for glucoamylase activity and R2 = 0.927 for biomass).

Lipase

A neural network of a relatively simple topology (4–7–1) with total carbon dioxide

Xylanase

The estimation and prediction of xylanase (1,4-β-xylan xylanohydrolase; EC 3.2.1.8) activity is an example of a stepwise learning procedure. The data discussed here were from 100 m3 industrial fed-batch fermentations, but this case also differs from the previous examples in that both the product and the production strain varied from one fermentation to another. Altogether, more than 20 different strains were used for the production of varying enzyme preparations. All the strains were

α-Amylase

The application of genetically modified microorganisms presents special difficulties in process control, because such organisms are often highly sensitive to the operating conditions. Fed-batch processes are particularly vulnerable to environmental changes. Consequently, convenient and reliable means of detecting process faults at an early stage could be of great value. Neural networks have been suggested for process-fault detection in a number of cases. Typically, one output neuron is assigned

β-Galactosidase

In industrial enzyme fermentations, the ability accurately to predict the end point of fermentation can be of great economic importance. During the production of β-galactosidase by an industrial autolytic strain of Streptococcus salivarius subsp. thermophilus 11F, the intracellular enzyme is rapidly released into the production medium upon cell lysis after a relatively short growth period51. In fermentations carried out in 3 l bioreactors, maximum enzyme activity was typically reached one to

Conclusions

Commercial applications of neural networks are becoming a reality, nearly half a century after the basic concept of neural networks was introduced in a search for ways to simulate the function of the brain. Today, for instance, neural networks have been applied to numerous every-day problems, such as controling the kerosene fan heaters widely used in Japan, and advanced hybrid neuro–fuzzy control systems are already in use in many household appliances, such as washing machines.

An example of

Acknowledgements

The authors are grateful to the Academy of Finland for financial support.

References (56)

  • T. Eerikäinen et al.

    Trends Food Sci. Technol.

    (1993)
  • A. Lübbert et al.

    Trends Biotechnol.

    (1994)
  • M.J. Willis et al.

    Automatica

    (1992)
  • G.A. Montague et al.

    Trends Biotechnol.

    (1994)
  • T. Eerikäinen et al.

    Food Control

    (1994)
  • Y-H. Zhu et al.

    Biochem. Eng. J.

    (1996)
  • K. Hornik et al.

    Neural Networks

    (1989)
  • S. Linko et al.

    J. Biotechnol.

    (1997)
  • P. Linko et al.

    Process Biochem.

    (1992)
  • P.R. Patnaik

    Process Biochem.

    (1997)
  • K. Hornik et al.

    Neural Networks

    (1989)
  • Y. Hirose et al.

    Neural Networks

    (1991)
  • K. Ye et al.

    J. Ferment. Biotechnol.

    (1994)
  • J. Glassey et al.

    J. Biotechnol.

    (1997)
  • V. Venkatasubramanian et al.

    Comput. Chem. Eng.

    (1990)
  • T. Eerikäinen et al.

    Food Control

    (1994)
  • O. Kirk et al.

    Tetrahedron Comput. Methodol.

    (1990)
  • Stroh, W. H. (1998) Genet. Eng. News 18 (5),...
  • K. Shimizu

    Adv. Biochem. Eng. Biotechnol.

    (1993)
  • S. Linko et al.

    Dev. Food Sci.

    (1994)
  • E.R. Crowe et al.

    Chem. Eng. Prog.

    (1995)
  • P. Linko et al.

    J. Biotechnol.

    (1991)
  • J. Thibault et al.

    Biotechnol. Bioeng.

    (1990)
  • M.N. Karim et al.

    Adv. Biochem. Eng. Biotechnol.

    (1993)
  • Samad, T. (1992) Proceedings of 1992 IEEE International Conference on Systems, Man, and Cybernetics, pp. 369–374, IEEE...
  • J.M. Benitez et al.

    Neural Networks

    (1997)
  • R.P. Lippmann

    IEEE Acoust. Speech Signal Process. Mag.

    (1987)
  • S.P. Chitra

    Chem. Eng. Prog.

    (1993)
  • Cited by (0)

    View full text