Identifying relevant molecular descriptors related to carcinogenic activity of polycyclic aromatic hydrocarbons (PAHs) using pattern recognition methods.

Abstract

Polycyclic Aromatic Hydrocarbons (PAHs) constitute an important family of molecules capable of inducing chemical carcinogenesis. In this work we report structure-activity relationship (SAR) studies for 81 PAHs using the pattern-recognition methods Principal Component Analysis (PCA), Hierarchical Clustering Analysis (HCA) and Neural Networks (NN). The used molecular descriptors were obtained from the semiempirical Parametric Method 3 (PM3) calculations. We have developed a new procedure that is capable of identifying the PAHs' carcinogenic activity with an accuracy higher than 80%. PCA selected molecular descriptors that can be directly correlated with some models proposed to PAHs' metabolic activation mechanism leading to the formation of PAHs-DNA adducts. PCA, HCA and NN validate the energy separation between the highest occupied molecular orbital and its next lower level as a major descriptor defining the carcinogenic activity. This descriptor has been only recently discussed in the literature as one new possible universal parameter for defining the biological activity of several classes of compounds.