Prediction of Acute Rodent Toxicity on the Basis of Chemical Structure and Physicochemical Similarity

A development of the Arithmetic Mean Toxicity (AMT) approach is presented in this article. Twenty six physicochemical descriptors, calculated by using the HYBOT program, along with molecular weight and lipophilicity were included in the selection of structural and physicochemical neighbours (analogues). Toxicity predictions of 906 chemicals from the REACH Pre-Registration Substance (PRS) list were carried out with the application of six nearest structural neighbours and three pairs of structural/physicochemical neighbours on the basis of molecular polarizability, the sum of negative atomic charges in a molecule, the sum of H-bond acceptor and donor factors and the octanol-water partition coefficient. The best prediction results were obtained three pair structural neighbours were applied (each pair contains one chemical with a higher and one chemical with a lower descriptor value). The prediction of toxicity as the mean arithmetic toxicity value of the nearest structural and physicochemical neighbours can be considered a robust approach for the read-across of properties (toxicity data) between analogues. Traditionally, analogues would be selected by expert judgement, but increasingly the availability of large databases and the application of nearest neighbour approaches such as the AMT approach provide a means of automating such assessments.