Avaliação de modelos QSAR em dados toxicológicos

Abstract

QSAR (Quantitative Structure Activity relationships) is an approach to pharmaceutical chemistry that is based on the assumption that the structure of a molecule (in its geometric, steric and electronic conjuncture) contains the characteristics responsible for its physical, chemical and chemical properties. and biological. Considering that this relationship can be described by means of numerical descriptors obtained from molecular modeling, it is logical to assume statistical models capable of inferring about the biological activity and chemical properties of organic molecules. The dataset studied presents as a dependent variable the logarithmic amount bioaccumulated in organic tissues of 779 chemical molecules, and the nine molecular descriptors were generated to predict their biological activity. With these data, through multiple linear regression, principal component regression and decision tree models were created to evaluate the biological activity, considering the entire process of validation and optimization of each model. The presented results demonstrated the different performances of the models within the data set and their comparison provided subsidies to evaluate them.