THE USE OF ARTIFICIAL-INTELLIGENCE SYSTEMS FOR PREDICTING TOXICITY

This review concentrates on the use of artificial intelligence (AI) systems for the prediction of toxic hazard via the establishment of structure-activity correlations. Methods for the analysis of the structure and physicochemical properties of molecules referred to include topological analysis, molecular orbital calculations, input of chemical structures, molecular modelling, cluster analysis and pattern recognition. Emphasis is placed on the importance of identifying substructural fragments of sufficient size and physicochemical specificity to act as toxicophores. Procedures for processing structural information include decision-tree and probabilistic systems, as well as algebraic and related statistical analyses for obtaining quantitative structure-activity relationships (QSARs). The principal differences between knowledge-based and automated rule-induction expert systems, and their utilisation for predicting the activity of chemicals, are discussed by reference to the use of several methods, including DEREK, HAZARDEXPERT, COMPACT, CASE and TOPKAT. It is concluded that these AI expert approaches have an important role to play in predictive toxicity screening as alternatives to animal experiments. Also, knowledge-based expert systems are being developed further for risk assessment.