In silico identification of endogenous and exogenous agonists of Estrogen-related receptor α

Estrogen-related receptor (ERR) α has recently caught the attention of toxicologists, since its gene expression in breast, prostate and endometrial cancers has a poor prognosis for these diseases. In this study, QSAR models for ERRα agonism and luciferase inhibition were developed for the in silico determination of potential ERRα agonists among common endogenous and exogenous substances from the Human Metabolome Database and EU Pesticides Database. The role of the luciferase inhibition model is to detect potential False Positives among predicted ERRα agonists due to interference with the assays reporter. An Artificial Bee Colony (ABC) algorithm was used to find optimal models parameters that are optimal for developing models with the best Balance Accuracy, Sensitivity, Specificity and Positive Predictive Value. Optimized models were merged to create combined ones in order to overcome the shortcomings of individual models. Combined models were used for prediction and a final prediction call scheme was designed to combine the ERRα agonism and luminescence inhibition predictions into a clear activity evaluation. 16 chemicals are recommended for ERRα agonism testing, most notable being isoflavones glycitein and naringenin, as well as, several metabolites of daidzein and genistein. The importance of the counterscreen assay and literature data in support of testing is discussed. Models were validated according to OECD principles and a QMRF report for combined ERRα agonism model was created. © 2019 Elsevier B.V.