Classification of hydroxylated polychlorinated biphenyls as agonists and nonagonists of estrogen receptors using linear discriminant analysis and decision tree models
Lukman Kehinde Akinola, Adamu Uzairu, Gideon Adamu Shallangwa, Stephen Eyije Abechi, Abdullahi B Umar · Environmental Toxicology and Chemistry · 2025
Hydroxylated polychlorinated biphenyls (OH-PCBs) are potential endocrine disruptors due to their interaction with nuclear receptors. However, experimental evaluation of their estrogenic activity is costly and time-consuming, limiting data availability. In this study, quantitative structure-activity relationship (QSAR) models were constructed using linear discriminant analysis (LDA) and decision tree (DT) with both 2D autocorrelation and arithmetic residuals in K-groups analysis (ARKA) descriptors to classify OH-PCBs as agonists or nonagonists of estrogen receptors (ERα and ERβ). For the ERα dataset, the training, test, and cross-validation set accuracies were 89.2%, 84.0%, and 88.0% for the LDA model developed with 2D autocorrelation descriptors (Model I); 89.2%, 72.0%, and 84.9% for the DT model developed with 2D autocorrelation descriptors (Model II); and 89.2%, 80.0%, and 87.0% for the ARKA-based model (Model V). Area under receiver operating characteristic (AUC-ROC) values of 0.959, 0.903, and 0.954 were obtained for Models I, II, and V respectively. For the ERβ dataset, the training, test, and cross-validation set accuracies were 90.5%, 84.0%, and 87.9% for the LDA model constructed with 2D autocorrelation descriptors (Model III); 89.2%, 68.0%, and 83.9% for the DT model constructed with 2D autocorrelation descriptors (Model IV); and 87.8%, 80.0%, and 84.9% for the ARKA-based model (Model VI). Values for AUC-ROC of 0.966, 0.892, and 0.945 were obtained for Models III, IV, and VI respectively. Overall, the QSAR models reported in this article provide a reliable and efficient approach for screening OH-PCBs for estrogenic activity, offering valuable tools for environmental risk assessment, with ARKA descriptors serving as effective alternatives to conventional descriptors.