Prediction of mutagenicity utilizing a hierarchical QSAR approach

Quantitative structure-toxicity (QSTR) models for the mutagenicity of a set of 95 aromatic amines were developed using four classes of calculated molecular descriptors, viz., topostructural, topochemical, geometrical and quantum chemical indices. Topochemical indices gave the best predictive model when the different classes of parameters were used separately. When hierarchical QSTRs were developed using all four classes of descriptors, there was a significant increase in explained variance by the addition of topochemical indices to the set of independent variables. The addition of geometrical and quantum chemical indices or log P to the set of descriptors resulted in very little improvement in model quality.