The Anti-Influenza Activity of Cyclohexene Derivatives Studied by Topological Indices and Neural Network

Abstract

In order to establish quantitative structure-activity relationship (QSAR) model of cyclohexene derivatives as neuraminidase inhibitors, the relationships between the anti-influenza activity of 34 cyclohexene derivatives and their molecular connectivity indices (mX) as well as their electrotopological state indices (Im) were analyzed, meanwhile molecular structures of the substances were also effectively characterized. On the one hand, two molecular connectivity 0X, 3X and four electrotopological state indices I1, I3, I6, I16, which were obtained by optimization of multiple linear stepwise regression, were employed to found QSAR model through stepwise regression analysis, where the correlation coefficient of equation was 0.906 and the average absolute error between experimental values and estimates was 0.37. On the other hand, Using these structural parameters above as the input neurons, a satisfying neural network model with the 6-2-1 network architecture was also constructed by the back-propagation (BP) algorithm, where the total correlation coefficient reached up to 0.966 and the average absolute error between experimental values and estimates of the model decreased to 0.19. Therefore it was concluded that the results of BP network were better than those of multiple linear regression methods.