Abstract

Coronavirus disease was one of the most challenging pandemics of the current century and it is necessary to design most efficient inhibitors for that. Quantitative Structure-Activity Relationship (QSAR) has a very important role in manufacturing novel and potent compounds for predicting the best new compounds activity. In this study, expert QSAR systems comprising of fuzzy inference system, neuro-fuzzy model and neural networks were developed for training inhibitors of SARS-CoV PLpro using 1558 descriptors. Six descriptors out of 1558 including RDF035u (Radial Distribution Function-035), RDF050u (Radial Distribution Function-050), Mor23u (signal 23/unweighted 3D-MoRSE descriptors), E2e (2nd component accessibility directional WHIM index/weighted by Sanderson electronegativity WHIM descriptors), R4p (R autocorrelation of lag 4/weighted by polarizability GETAWAY descriptors) and B04[N-N] (Presence/ absence of N-N at topological distance 4 2D Atom Pairs) were applied as the optimal inputs for estimating pIC50 (-log[IC₅₀]) of PLpro inhibitors. Using the weighted averaging committee machine, the correlation coefficient increased to 0.86 that corresponds to the mean squared error of 0.121. The results of this study enable researchers to predict the behaviour and activity of new molecules and compounds based on the behaviour of similar molecules to control SARS-CoV-2. The advanced committee machine models can successfully be employed in drug discovery and design.

Keywords:Quantitative structure-activity relationship; Papain-like protease; Coronavirus; Committee machine; Neural network; Descriptors; Fuzzy inference system; Neuro-fuzzy model