Molecular level studies on the origin of toxicity: Identification of key variables and selection of descriptors

A study of toxicity at the molecular level is repotted. Twenty key variables or descriptors were identified using regression and principal component analysis as determinants of relevance. Most of the key descriptors described various size and electronic characteristics tentatively identified as being associated with toxicity. Other key descriptors consisted of complex structures associated with ortho-substitution, which were also suspected to substantially increase toxicity. A model employing the twenty key variables identified was developed using neural networks (NNs). The resulting model demonstrated good predictive capability when tested on compounds not in the training set.