Activity prediction of hepatitis C virus NS5B polymerase inhibitors of pyridazinone derivatives

• The QSAR models of inhibitory activity of pyridazinone derivatives are constructed.
• PLS and PSO-SVM are used to construct the models.
• The important molecular descriptors are selected by stepwise-MLR and UVE-PLS.
• The performance of PSO-SVM model is more accurate than PLS model.

A valid quantitative structure–activity relationship (QSAR) model was applied to predict IC50 value of pyridazinone derivatives as HCV NS5B protease inhibitors. Various chemical descriptors were calculated by E-Dragon. Six character variables were selected though stepwise multiple linear regression (stepwise-MLR), which included MATS6m, RDF055e, Mor31u, G3m, R1m and R4v +. In addition, twenty-three molecular descriptors were obtained via uninformative variable elimination by partial least squares (UVE-PLS). The selected descriptors using two approaches were basically the same type of molecular descriptors. Subsequently, partial least squares (PLS) and particle swarm optimization support vector machine (PSO-SVM) were utilized to establish the linear and nonlinear models by two set of descriptors and their activity data, respectively. The predictive performance of the proposed models was evaluated by the strict criteria. The results showed that the predictive power of the PSO-SVM models was better than the corresponding PLS models. Thus, it can be inferred that the PSO-SVM models were robust and satisfactory, and could provide some feasible and effective information to design and synthesis of highly potent HCV NS5B polymerase inhibitors.