Three modelling systems (MultiCase (R), LeadScope (R) and MDL (R) QSAR) were used for construction of androgenic receptor antagonist models. There were 923-942 chemicals in the training sets. The models were cross-validated (leave-groups-out) with concordances of 77-81%, specificity of 78-91% and sensitivity of 51-76%. The specificity was highest in the MultiCase (R) model and the sensitivity was highest in the MDL (R) QSAR model. A complementary use of the models may be a valuable tool when optimizing the prediction of chemicals for androgenic receptor antagonism. When evaluating the fitness of the model for a particular application, balance of training sets, domain definition, and cut-offs for prediction interpretation should also be taken into account. Different descriptors in the modelling systems are illustrated with hydroxyflutamide and dexamethasone as examples (a non-steroid and a steroid anti-androgen, respectively). More research concerning the mechanism of anti-androgens would increase the possibility for further optimization of the QSAR models. Further expansion of the basis for the models is in progress, including the addition of more drugs.
Jensen, G. E., Nikolov, N. G., Wedebye, E. B., Ringsted, T., & Niemelä, J. R. (2011). QSAR models for anti-androgenic effect - a preliminary study. S A R and Q S A R in Environmental Research, 22(1-2, Sp. Iss. SI), 35-49. https://doi.org/10.1080/1062936X.2010.528981