The HQSAR, molecular docking, and ROCS were applied to a data-set of 57 cruzain inhibitors. The best HQSAR model (q2 = .70, r2 = .95, [Formula: see text] = .62, [Formula: see text] = .09 and [Formula: see text] = .26), employing well-balanced, diverse training (40) and test (17) sets, was obtained using atoms (A), bonds (B), and hydrogen (H) as fragment distinctions and 6-9 as fragment sizes. This model was then used to predict the unknown potencies of 121 compounds (the V1 database), giving rise to a satisfactory predictive r2 value of .65 (external validation). By employing an extra external data-set comprising 1223 compounds (the V3 database) either retrieved from the ChEMBL or CDD databases, an overall ROC AUC score well over .70 was obtained. The contribution maps obtained with the best HQSAR model (model 3.4) are in agreement with the predicted binding mode and with the biological potencies of the studied compounds. We also screened these compounds using the ROCS method, a Gaussian-shape volume filter able to identify quickly the shapes that match a query molecule. The area under the curve (AUC) obtained with the ROC curves (ROC AUC) was .72, indicating that the method was very efficient in distinguishing between active and inactive cruzain inhibitors. These set of information guided us to propose novel cruzain inhibitors to be synthesized. Then, the best HQSAR model obtained was used to predict the pIC50 values of these new compounds. Some compounds identified using this method have shown calculated potencies higher than those which have originated them.
Keywords: HQSAR; ROCS; cruzain and nitrile-containing inhibitor; molecular docking.