Comparing machine learning models for acetylcholine esterase inhibitors

Acetylcholinesterase is the main neurotransmitter in the cholinergic system. Impairment of the cholinergic system can be a reason for Alzheimer's and multiple sclerosis. Alzheimer's disease and multiple sclerosis affect patients and their relatives' daily lives enormously. New therapies with more benefits than current therapies for these diseases would facilitate patients' lives. In this respect, discovering novel acetylcholine esterase inhibitors with more effective and fewer side effects is highly important.  Machine learning algorithms are very useful to predict the activity of molecules for a biological target. In this study, our classification models were built with Deep Neural Networks (DNN), Support Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost) to predict molecules as active or inactive for acetylcholinesterase inhibitors. These models were evaluated with various metrics. As a result, The DNN model showed a better ability to classify (accuracy=0.93, F1 score=0.88, MCC=0.8, Roc-Auc=0.89 in the test set) molecules than the other models.