Maciej Szaleniec1. 1. Institute of Catalysis and Surface Chemistry, Polish Academy of Science, Kraków, Poland. ncszalen@cyfronet.pl
Abstract
BACKGROUND: Artificial Neural Networks (ANNs) are introduced as robust and versatile tools in quantitative structure-activity relationship (QSAR) modeling. Their application to the modeling of enzyme reactivity is discussed, along with methodological issues. Methods of input variable selection, optimization of network internal structure, data set division and model validation are discussed. The application of ANNs in the modeling of enzyme activity over the last 20 years is briefly recounted. METHODS: The discussed methodology is exemplified by the case of ethylbenzene dehydrogenase (EBDH). Intelligent Problem Solver and genetic algorithms are applied for input vector selection, whereas k-means clustering is used to partition the data into training and test cases. RESULTS: The obtained models exhibit high correlation between the predicted and experimental values (R(2) > 0.9). Sensitivity analyses and study of the response curves are used as tools for the physicochemical interpretation of the models in terms of the EBDH reaction mechanism. CONCLUSIONS: Neural networks are shown to be a versatile tool for the construction of robust QSAR models that can be applied to a range of aspects important in drug design and the prediction of biological activity.
BACKGROUND: Artificial Neural Networks (ANNs) are introduced as robust and versatile tools in quantitative structure-activity relationship (QSAR) modeling. Their application to the modeling of enzyme reactivity is discussed, along with methodological issues. Methods of input variable selection, optimization of network internal structure, data set division and model validation are discussed. The application of ANNs in the modeling of enzyme activity over the last 20 years is briefly recounted. METHODS: The discussed methodology is exemplified by the case of ethylbenzene dehydrogenase (EBDH). Intelligent Problem Solver and genetic algorithms are applied for input vector selection, whereas k-means clustering is used to partition the data into training and test cases. RESULTS: The obtained models exhibit high correlation between the predicted and experimental values (R(2) > 0.9). Sensitivity analyses and study of the response curves are used as tools for the physicochemical interpretation of the models in terms of the EBDH reaction mechanism. CONCLUSIONS: Neural networks are shown to be a versatile tool for the construction of robust QSAR models that can be applied to a range of aspects important in drug design and the prediction of biological activity.
Authors: A Dudzik; W Snoch; P Borowiecki; J Opalinska-Piskorz; M Witko; J Heider; M Szaleniec Journal: Appl Microbiol Biotechnol Date: 2014-12-31 Impact factor: 4.813