QSAR classification-based virtual screening followed by molecular docking studies for identification of potential inhibitors of 5-lipoxygenase.

Developments of novel inhibitors to prevent the function of 5-lipoxygenase (5-LOX) proteins that are responsible for a variety of inflammatory and allergic disease are a major challenge in the scientific community. In this study, robust QSAR classification models for predicting 5-LOX activity were developed using machine learning algorithms. The Support Vector Machines (SVM), Logistic Regression, k-Nearest Neighbour (NN) and Decision Trees were adopted to improve the prediction ability of the classification models. The most informative molecular descriptors that contribute to the prediction of 5-LOX activity are screened from e-Dragon, Ochem, PowerMV and Combined databases using Filter-based feature selection methods such as Correlation Feature Selection (CFS) and Information Gain (IG). Performances of the models were measured by 5-fold cross-validation and external test sets prediction. Evaluation of performance of feature selection revealed that the CFS method outperforms the IG method for all descriptor databases except for PowerMV database. The best ensemble classification model was obtained with the IG filtered 'PowerMV' descriptor database using kNN (k = 5) algorithm which displayed an overall accuracy of 76.6% for the training set and 77.9% for the test set. Finally, we employed this model as a virtual screening tool for identifying potential 5-LOX inhibitors from the e-Drug3D drug database and found 43 potential hit candidates. This top screened hits containing one known 5-LOX inhibitors zileuton as well as novel scaffolds. These compounds further screened by applying molecular docking simulation and identified four potential hits such as Belinostat, Masoprocol, Mefloquine and Sitagliptin having a comparable binding affinity to zileuton.