Abstract
Current challenges in drug designing and lead optimization has reached a bottle neck where the main onus lies on rigorous validation to afford robust and predictive models. In the present study, we have suggested that predictive structure‐activity relationship (SAR) models based on robust statistical analyses can serve as effective screening tools for large volume of compounds present either in chemical databases or in virtual libraries. 3D descriptors derived from the similarity‐based alignment of molecules with respect to group center overlap from each individual template point and other “alignment averaged,” but significant descriptors (ClogP, molar refractivity, connolly accessible area) were used to generate QSAR models. The results indicated that the artificial neural network method (r^2^ = 0.902) proved to be superior to the multiple linear regression method (r^2^ = 0.810). Cross validation of the models with an external set was reasonably satisfactory. Screening PubChem compound database based on the models obtained, yielded 14 newer modified compounds belonging to the TIBO class of inhibitors, as well as, two novel scaffolds, with enhanced binding efficacy as hits. These hits may be targeted toward potent lead‐optimization and help in designing and synthesizing new compounds with potential therapeutic utility. © 2008 Wiley Periodicals, Inc. J Comput Chem, 2009