Comparison Accuracy of Multi-Layer Perceptron and DNN in QSAR Classification for Acetylcholinesterase Inhibitors

Recently, the development of the artificial intelligence approach is a solution for evaluating the effectiveness, analysis, and safety of drug candidates due to a large number of data sets available. One of the approaches to artificial intelligence is deep learning. Deep learning has a significant influence on drug discovery procedures for rational drug development and optimization so that it can affect public health. The discovery of various inhibitors needs reliable models to figure out the side effects of the drug without requiring large costs and long amounts of time. A new way for the treatment of Alzheimer's disease is Acetylcholinesterase inhibitors. The Quantitative Structure-Activity Relationship (QSAR) model is a model used to filter large databases of the compound to figure the biological properties of chemical molecules based on their structure. The modeling that was used in this study was QSAR classification. The QSAR classification model predicted active and inactive compounds in Acetylcholinesterase inhibitors. There were 3809 inhibitors which consisted of 2215 active inhibitors and 1594 inactive inhibitors. They were labeled using fingerprints as descriptors. This study compared the performances of MLP and DNN in the classification. The result of this study showed DNN had better accuracy of 0.841 in classification.