Design, synthesize and antiurease activity of novel thiazole derivatives: Machine learning, molecular docking and biological investigation

Abstract

Machine learning is one of the methods used in the design of new molecules with different biological properties and has become a trend in recent years. Since there are many published studies on urease enzyme inhibition, we accumulated a huge literatüre data set and improved a model for antiurease activity. The balanced accuracy of the selected compounds (classification models) were about 78% and the predictive accuracy of them possessed a coefficient of determination q2 = 0.2-0.7 (regression models) with cross-validation and independent test sets. Thanks to the chemical library created with the machine learning method, a comparison of the predictive and experimental results of the compounds that previously synthesized by us and investigated urease inhibition was made. Compounds observed to be experimentally active were found to be active with the machine learning method. The models are freely avaible online (http://ochem.eu) and can be used to predict potantial antiurease activity of novel compounds. The activity potentials of compounds 4a-d were further evaluated via molecular docking studies with AutoDock4 and AutoDock Vina softwares.