Prediction of COMT Inhibitors Using Machine Learning and Molecular Dynamics Methods.

Catechol -methyltransferase (COMT) plays a vital role in deactivating neurotransmitters like dopamine, norepinephrine, etc., by methylating those compounds. However, the deactivation of an excess amount of neurotransmitters leads to serious mental ailments such as Parkinson's disease. Molecules that bind inside the enzyme's active site inhibit this methylation mechanism by methylating themselves, termed COMT inhibitors. Our study is focused on designing these inhibitors by various machine learning methods. First, we have developed a classification model with experimentally available COMT inhibitors, which helped us generate a new data set of small inhibitor-like molecules. Then, to predict the activity of the new molecules, we have applied regression techniques such as Random Forest, AdaBoost, gradient boosting, and support vector machines. Each of the regression models yielded an value > 70% for both training and test data sets. Finally, to validate our models, 200 ns long molecular dynamics (MD) simulations of the two known inhibitors with known IC values and the resultant inhibitors were performed inside the binding pockets to check their stability within. The free energy barrier of the methyl transfer from -adenosyl-l-methionine (SAM) to each inhibitor was determined by combining steered molecular dynamics (SMD) and umbrella sampling using the quantum mechanics/molecular mechanics (QM/MM) method.