Calculations of acute intravenous toxicity in mice based on local regression models in superoverlapping clusters (LRMSC)

Modeling of quantitative structure — activity relationships (QSAR) between physicochemical descriptors of organic chemicals and their acute intravenous toxicity in mice have been presented. This approach includes three steps: structure-similarity chemicals selection for every compound-of-interest (clusterization); construction of quantitative structure — toxicity models for every cluster (without including of compounds-of-interest); application of the obtained QSAR equations for chemical-of-interest toxicity estimation. This approach has been applied for calculations of acute intravenous toxicity for 10241 organic chemicals. For 7759 compounds possessing structural neighbors with the Tanimoto index (Tc) of 0.30 and above the standard deviation of the calculated vs. experimental log(1/LD 50 ) values was 0.51 at the estimation of the experimental determination error of ±0.50 (log(1/LD 50 ) value). Calculations performed for remaining compounds (∼24%) were not as good as those made for the former group, possibly due to lack of reasonable number of structurally related analogues. It’s suggested that this QSAR approach can be useful for prediction of biological activity and toxicity of large sets of chemical compounds.