Predictive QSAR modeling of phosphodiesterase 4 inhibitors

A series of diverse organic compounds, phosphodiesterase type 4 (PDE-4) inhibitors, have been modeled using a QSAR-based approach. 48 QSAR models were compared by following the same procedure with different combinations of descriptors and machine learning methods. QSAR methodologies used random forests and associative neural networks. The predictive ability of the models was tested through leave-one-out cross-validation, giving a Q2 = 0.66–0.78 for regression models and total accuracies Ac = 0.85–0.91 for classification models. Predictions for the external evaluation sets obtained accuracies in the range of 0.82–0.88 (for active/inactive classifications) and Q2 = 0.62–0.76 for regressions. The method showed itself to be a potential tool for estimation of IC50 of new drug-like candidates at early stages of drug development.

Graphical abstractFigure optionsDownload full-size imageDownload high-quality image (263 K)Download as PowerPoint slideHighlights► We have presented series of new predictive QSAR models. ► Data set was consisted of 1015 phosphodiesterase-4 inhibitors. ► QSAR methodologies used random forests and neural networks. ► The models demonstrated a good predictive ability. ► The method to be a potential tool for estimation of new drug-like candidates.