Good and bad molecular fingerprints for human rhinovirus 3C protease inhibition: identification, validation, and application in designing of new inhibitors through Monte Carlo-based QSAR study

Abstract

HRV 3 C protease (HRV 3Cpro) is an important target for common cold and upper respiratory tract infection. Keeping in view of the non-availability of drug for the treatment, newer computer-based modelling strategies should be applied to rationalize the process of antiviral drug discovery in order to decrease the valuable time and huge expenditure of the process. The present work demonstrates a structure wise optimization using Monte Carlo-based QSAR method that decomposes ligand compounds (in SMILES format) into several molecular fingerprints/descriptors. The current state-of-the-art in QSAR study involves the balance of correlation approach using four different sets: training, invisible training, calibration, and validation. The final models were also validated through mean absolute error, index of ideality of correlation, Y-randomization and applicability domain analysis. R2 and Q2 values for the best model were 0.8602, 0.8507 (training); 0.8435, 0.8331 (invisible training); 0.7424, 0.7020 (calibration); 0.5993, 0.5216 (validation), respectively. The process identified some molecular substructures as good and bad fingerprints depending on their effect to increase or decrease the HRV 3Cpro inhibition. Finally, new inhibitors were designed based on the fundamental concept to replace the bad fragments with the good fragments as well as including more good fragments into the structure. The study points out the importance of the fingerprint based drug design strategy through Monte Carlo optimization method in the modelling of HRV 3Cpro inhibitors.