Abstract
Neuraminidase (NA) is a glycoprotein found on the surface of influenza A virus that is used for releasing new progeny of virions by cleaving the terminal sialic acid residue from the surface of infected cells. Therefore, NA is an interesting potential target to design promising NA inhibitiors to serve as antiviral agents for preventing viral propagation. In this study, a data set of 61 H1N1 neuraminidase inhibitors of influenza A virus was employed in the construction of quantitative structure-activity relationship (QSAR) model using the CORrelation And Logic (CORAL) software available at http://www.insilico.eu/coral. The chemical structure of compounds in the SMILES format was used as input to CORAL in discerning the correlation between an endpoint (i.e. neuraminidase inhibitory activity) and their corresponding molecular descriptors. Three random splits of the data into sub-training, calibration and testing sets were carried out. The optimal threshold and number of epoch to use in building the QSAR models were derived from the CORAL software. Results indicated that the QSAR models displayed good prediction performance as deduced from statistical parameters affording r2 = 0.7783-0.9166, 0.7609-0.8336 and 0.8384-0.9069 and q2 = 0.7453-0.8924, and 0.7311-0.7939 and 0.8051-0.8843 for sub-training, calibration and test set, respectively. Furthermore, F value and standard error of estimation provided good statistical results for the predictive performance of QSAR models. Interpretations of the derived structure-activity relationship provided pertinent knowledge on the origins of good and poor neuraminidase inhibitory activity. Therefore, the QSAR model holds great potential for the rational design of novel neuraminidase inhibitors.
Keywords: Neuraminidase inhibitors, Influenza virus, SMILES, CORAL, QSAR, Data mining.
Graphical Abstract