|
(83-12) A Neuro-Fuzzy Approach of Phenols Toxicity Prediction. NEAGU, Ciprian-Daniel*,1, APTULA, Aynur2, GINI, Giuseppina1, CRONIN, Mark3, NETZEVA, Tatjana3, 1 Dipartimento di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci 32, Milano, Italy2 Department of Chemical Ecotoxicology, UFZ Centre for Environmental Research, Leipzig-Halle, Permoserstrasse 15, Leipzig, Germany3 School of Pharmacy and Chemistry, Liverpool John Moores University, Byrom Street, Liverpool, England ABSTRACT- In recent years, the neuro-fuzzy systems have drawn increasing research interest. This approach has been successfully used in various areas, such as natural language understanding, robotics, medical diagnosis, particularly based on neuro-fuzzy networks. The aim of this study was to model toxicity to Tetrahymena pyriformis of 225 phenols, which had previously been classified into 5 groups, with the original developed system NIKE: Neural explicit&Implicit Knowledge inference systEm. The contribution of each descriptor for their influence in toxicity and mode of action (MOA) variations was measured and compared to the specific QSARs. Consequently, models for the reduced data sets were developed. The most spectacular result is the descriptors significance study. The neural models are more sensible to the noisy data, which make them a very important indicator of the significance of the descriptors to toxicity and MOA correlation. Artificial neural networks (ANNs) demonstrates specific characteristics about missing descriptors: a translation of the predictions, which announce a linear dependence with the absent descriptor, and a proportional magnification of the error, consequence of a nonlinear relation between some of the current input descriptors. The neuro-fuzzy models are more robust to noisy data than crisp neural networks or QSAR models. This conclusion recommends it as suitable for toxicity prediction task. This work was supported in part by the European Union IMAGETOX Research Training Network (HPRN-CT-1999-00015). Key words: Phenols Toxicity, Neuro-Fuzzy Networks, QSAR, Fuzzy Inference |
