Estimation of cytotoxicity of organic chemicals by Quantitave Structure Activity Relationships (QSARs) based on a Neural Network adaptation

Abstract

Recently, various kinds of chemicals have been produced and released to environment by human activities, and they caused various environmental problems. Therefore, environmental risks of chemicals should easily and quickly be examined. In this study, neural network was developed to predict cytotoxicity of human liver cells. Cytotoxicity data of 57 chemicals investigated by using in vitro cell survival assay performed by human liver cells: HepG2 was used for training neural networks. As input data for the neural network partition coefficient octanol-water (logPow), dissociation constant (pKa), molecular weight (M.W.), and concentration of the chemicals were used. Resultant cell survival data were derived on neural networks in a personal computer. In addition, the chemicals were classified into some groups in terms of chemical structure. Leave-one-out test was performed to validate the correlation between data experimental and predicted by the neural network for each chemical groups. There was a significant correlation between cell survival data experimental and predicted. As classified chemicals into smaller group according to the chemical property and structure can estimate the toxicity quantitatively in dose response manner.