Prediction of genotoxicity of various environmental pollutants by artificial neural network simulation

Abstract

In order to evaluate human carcinogenic risks, genotoxicity data such as animal cancer bioassay are often not available. In this study, to assess the relevance of indicator of carcinogenic risks, we used the "molecular diversity approach" to estimate the genotoxicity based upon Salmonella genotoxicity test using the umu test and systemic toxicity data of the 82 environmental chemicals predicted by neural network simulation. The 82 environmental chemicals were randomly selected for this study according to the production and usage in Japan. Even in this challenging trial for QSTR (Quantitative Structure Toxicity Relationship) study, approaches using artificial neural networks can account for about 94% of the variation in the genotoxicity results derived by the umu-test.