engleski

Toxicity of benzene and its derivatives toward mammals: development and applications of QSAR models

The goal of the study was to develop reliable models that could be used for prediction of toxicity (in vivo) of varied aromatic compounds toward mammals. Compounds used as training, and test groups were derivatives of benzene (single benzene ring with different substitute groups such as hydroxyl-, nitro-, amino-, methyl-, methoxy-, etc, ). A Genetic Algorithm and multiple regression analysis were applied to select the descriptors and to generate the correlation models. Evaluation of models was performed by calculating and comparing their model performances (r, R2, s, F, Q2) after splitting set of organic compounds to training and test sets. As the most predictive model is shown the 3-variable model having also a good ratio of the number of descriptors and their predictive ability to avoid overfitting. The main contribution to the toxicity showed MATS2p and C-026 descriptors, belonging to 2D autocorrelation and atom-centered fragments descriptors, respectively. The GA-MLRA approach showed good results in this study, which allows building simple, interpretable and transparent models that could be used for future studies of predicting toxicity of organic compounds to mammals.

QSPR; toxicity; aromatics; modeling

nije evidentirano