Abstract
Chmical function based pharmacophore models were developed for a series of pyrimidinone- and pyrazinone-based HCV NS3 protease inhibitors. The pharmacophore models were generated using a training set consisting of 20 inhibitors. The activity spread, expressed in IC50 of training set molecules was from 20 to 30000 nM. The most predictive pharmacophore model (hypothesis 1), consisting of three features, namely, two hydrophobic, one hydrogen bond donor and one hydrophobic aromatic, had a correlation (r) of 0.943 and a root mean square of 0.886, and the cost difference between null cost and fixed cost was 52.33 bits and the cost difference between null cost and total cost was 42.81 bits. The model was cross validated by randomizing the data using the CatScramble technique. The results confirmed that the pharmacophore models generated from the training set were not due to chance correlation. The best model (hypothesis 1) was validated using test set molecules (total of 30) and performed not bad in classifying active and inactive molecules, it is 76.67% success. The model was further validated by mapping onto it a diverse set of two HCV NS3 protease inhibitor identified by two different pharmaceutical companies. The best model predicted one compounds as being highly active and one inactive, maybe these two compounds work by different mechanism. These multiple validation approaches provide confidence in the utility of this pharmacophore model developed in this study as a 3D query tool in virtual screening to retrieve new chemical entities as potent HCV NS3 inhibitors.