In silico prediction of the inhibition of new molecules on SARS-CoV-2 3CL protease by using QSAR: PSOSVR approach

Continuous effort is dedicated to clinically and computationally discovering potential drugs for the novel coronavirus-2. Computer-Aided Drug Design CADD is the backbone of drug discovery, and shifting to computational approaches has become necessary. Quantitative Structure–Activity Relationship QSAR is a widely used approach in predicting the activity of potential molecules and is an early step in drug discovery. 3-chymotrypsin-like-proteinase 3CLpro is a highly conserved enzyme in the coronaviruses characterized by its role in the viral replication cycle. Despite the existence of various vaccines, the development of a new drug for SARS-CoV-2 is a necessity to provide cures to patients. In the pursuit of exploring new potential 3CLpro SARS-CoV-2 inhibitors and contributing to the existing literature, this work opted to build and compare three models of QSAR to correlate between the molecules’ structure and their activity: IC50 through the application of Multiple Linear Regression(MLR), Support Vector Regression(SVR), and Particle Swarm Optimization-SVR algorithms (PSO-SVR). The database contains 71 novel derivatives of ML300which have proven nanomolar activity against the 3CLpro enzyme, and the GA algorithm obtained the representative descriptors. The built models were plotted and compared following various internal and external validation criteria, and applicability domains for each model were determined. The results demonstrated that the PSO-SVR model performed best in predictive ability and robustness, followed by SVR and MLR. These results also suggest that the branching degree 6 had a strong negative impact, while the moment of inertia X/Z ratio, the fraction of rotatable bonds, autocorrelation ATSm2, Keirshape2, and weighted path of length 2 positively impacted the activity. These outcomes prove that the PSO-SVR model is robust and concrete and paves the way for its prediction abilities for future screening of more significant inhibitors' datasets.