Computer methods for the assessment of toxicity.

The prediction of the biological activity of chemical compounds by means of mathematical models is discussed. Biological activity of chemicals, including their toxicity on man and other biological organisms and systems, involves too complex phenomena to presently be predictable by fundamental models such as ab initio quantum mechanical models or statistical mechanical models. Hence one takes recourse to semi-empirical and empirical models which relate the variation in chemical structure of chemical compounds to the variation in their measured biological activity, e.g. toxicity in one or several test systems. These models are "calibrated" on series of similar compounds with "known" toxicity, the training set. Thereafter the models can--in fortunate cases--be used to predict the toxicity of compounds which are structurally similar to the training set compounds. The formulation and applicability of semi-empirical and empirical models relating chemical structure to biological activity is discussed. Causes and remedies for commonly encountered fallacies are presented.