Prediction of acute toxicity to mice by the Arithmetic Mean Toxicity (AMT) modelling approach

A modelling approach based on the structural and physicochemical similarity of chemicals to their nearest neighbours is proposed for toxicity estimation. This approach, called Arithmetic Mean Toxicity (AMT) modelling, is illustrated by means of an AMT model for predicting acute rodent toxicity. The AMT approach uses one or a few pairs of nearest structural neighbours. Each pair contains a chemical with a higher descriptor value and with a smaller descriptor value compared with the chemical of interest. Arithmetic mean toxicity values of those pairs are considered as toxicity of chemical of interest. The toxicity of the chemical of interest was not included in the development of the AMT model. The approach was applied to calculate the toxicity of chemicals to mice following intravenous injection. A toxicity data set containing 10,241 organic neutral compounds was formed from the SYMYX Toxicity database. The toxicity (log (1/LD(50)), mmol/kg), where LD(50) is the median lethal dose, of 10,227 chemicals was calculated with a standard deviation +/-0.52. A cascade AMT model was applied to estimate error values in calculations of toxicity of chemicals having different number structural neighbours and level of similarity. It was found that 7085 chemicals (about 69% of all chemicals in the data set) were calculated with a standard deviation in the interval (+/-0.33)-(+/-0.48), which is comparable to the experimental error of determination. For the remaining 3142 chemicals (about 31% of the data set), the standard deviation was +/-0.64. In the regulatory assessment of chemicals, the AMT approach could be used as a means of filling data gaps when experimental data are missing.