Perspectives on the past, present and future of computer prediction in toxicology

Prediction of toxic effects from chemical structure has a long history. The first QSAR (quantitative structure activity relationship) procedures in toxicology took place in the 1960s but as early as in the 1930s Hammett made a QSAR model for chemical reactivity. A better known model in biology is the Hansch approach based on a QSAR algorithm combining the aspects of physical molecular structure, solubility in biological membranes and chemical reactivity. This latter model has been used to predict biological effects of chemicals in homologous series of chemicals with considerable success, when the mechanism of action is common. The above approaches use mathematical models to quantify likely biological activity, hence `quantitative structure-activity relationships' which are most successful within relatively limited ranges of known chemistry. A broader approach is SAR (structure activity relationship), where specific molecular aspects of a chemical are used to predict qualitatively its possible biological action. Both the QSAR and the SAR approaches can be performed without computer power, an expert with the appropriate information and a calculator could come to the same prediction (although very much slower in most cases). However, the logical conclusion of SAR is to look at three-dimensional structures of molecules and their interaction with three-dimensional receptors, in all their most probable conformational states, for this procedure computing power is essential.