Prediction of toxicity and data exploratory analysis of estrogen-active endocrine disruptors using counter-propagation artificial neural networks

In this work, a novel algorithm for optimization of counter-propagation artificial neural networks has been used for development of quantitative structure–activity relationships model for prediction of the estrogenic activity of endocrine-disrupting chemicals. The search for the best model was performed using genetic algorithms. Genetic algorithms were used not only for selection of the most suitable descriptors for modeling, but also for automatic adjustment of their relative importance. Using our recently developed algorithm for automatic adjustment of the relative importance of the input variables, we have developed simple models with very good generalization performances using only few interpretable descriptors. One of the developed models is in details discussed in this article. The simplicity of the chosen descriptors and their relative importance for this model helped us in performing a detailed data exploratory analysis which gave us an insight in the structural features required for the activity of the estrogenic endocrine-disrupting chemicals.

Graphical abstractFigure optionsDownload full-size imageDownload high-quality image (338 K)Download as PowerPoint slideResearch highlights▶ Artificial neural networks (ANN) in the most cases are used as a black-box modeling algorithms. The disadvantage of the use of black-box modeling algorithms is that it is not possible to interpret how the independent variables influence on the dependent one. In the case of counter-propagation artificial neural networks (CPANN), due to the fact that different weigh levels do not interact directly (as in the back-propagation ANN), it is possible to evaluate the influence of the independent variables on the modeled property (1) if the number of independent variables is small and (2) if the independent variables (in QSAR: descriptors) are interpretable. ▶ In this work we tried to develop simple and interpretable model for prediction of toxicity of estrogen-active endocrine disruptors based on our novel algorithm for automatic adjustment of the relative importance of the input variables for optimization of CPANN. ▶ The algorithm is based on genetic algorithms which, in this case, are not used only for variable selection but also for automatic adjustment of the relative importance of the input variables. The fact that we were able to adjust the relative importance of the input variables helped us to develop simpler and interpretable models (with only few descriptors). The discussed model in the manuscript has good generalization performances (checked using external validation). These characteristics of the presented model gave us an opportunity, using data exploratory analysis, to get an insight into how different structural features of the substances in our (diverse) dataset influence their activity as endocrine disruptors.