Abstract
Recently, it is globally necessary to collect data on toxicities of numerous chemical substances available in market. But safety tests using animals take huge cost and period, and it is impossible to get toxicity data on all unascertained chemical substances by animal test. Also there are many problems in existing databases which collect toxicity data on chemical substances. As a screening method for animal tests, a toxicity prediction on a basis of quantitative structure-activity relationship (QSAR) is meaningful. Several toxicity prediction systems with QSAR have been developed, but the performances of most existing systems are highly questionable. In order to develop a toxicity prediction system for numerous chemical substances showing a higher performance than those of existing systems, we are applying a neural network technique. Several problems on toxicity information and its QSAR prediction of chemical substances are discussed.
