Discovery of Influenza A virus neuraminidase inhibitors using support vector machine and Naïve Bayesian models.

Neuraminidase (NA) is a critical enzyme in the life cycle of influenza virus, which is known as a successful paradigm in the design of anti-influenza agents. However, to date there are no classification models for the virtual screening of NA inhibitors. In this work, we built support vector machine and Naïve Bayesian models of NA inhibitors and non-inhibitors, with different ratios of active-to-inactive compounds in the training set and different molecular descriptors. Four models with sensitivity or Matthews correlation coefficients greater than 0.9 were chosen to predict the NA inhibitory activities of 15,600 compounds in our in-house database. We combined the results of four optimal models and selected 60 representative compounds to assess their NA inhibitory profiles in vitro. Nine NA inhibitors were identified, five of which were oseltamivir derivatives with large C-5 substituents exhibiting potent inhibition against H1N1 NA with $$\hbox {IC}_{50}$$ values in the range of 12.9-185.0 nM, and against H3N2 NA with $$\hbox {IC}_{50}$$ values between 18.9 and 366.1 nM. The other four active compounds belonged to novel scaffolds, with $$\hbox {IC}_{50}$$ values ranging 39.5-63.8 $$\upmu $$ M against H1N1 NA and 44.5-114.1 $$\upmu $$ M against H3N2 NA. This is the first time that classification models of NA inhibitors and non-inhibitors are built and their prediction results validated experimentally using in vitro assays. [ABSTRACT FROM AUTHOR]