An Automated Approach for Prediction of Half-Maximal Inhibitory Concentration (pIC50) Values, Bioactivity Classes of Acetylcholinesterase Inhibitors and Its Application to Alzheimer’s Disease

In the drug discovery field, identifying the potential drug molecule for a target involves enormous time and cost for scientists. Approaches like Quantitative Structure Activity Relationships (QSAR) help in the identification of the relationships between the physicochemical properties of the compounds and their associated biological activities using statistical models that are facilitates to predict the biological responses of new drug-like molecules which saves time and cost in the drug development process. This research focus on the half-maximal inhibitory concentration (pIC50) values prediction using a machine learning model for the drug molecules against the acetylcholinesterase (AChE) which is an enzyme that is responsible for the acetylcholine degradation by the cleavage of ester bonds. This is the most vital drug target majorly used for Alzheimer’s disease treatment. The Inhibition of this enzyme increases the availability of acetylcholine in the synaptic clef that helps to escalate the cognitive ability in humans. The ChEMBL dataset contains the IC50 values for AChE drug molecules and 12 different fingerprints are generated using PaDEL software that is used to develop and compare the evaluation metrics of different regression and multi-class classification models. A comparison study of the different models with each fingerprint, class imbalance, and hyperparameter tuning is handled in this study. Also, the output will be given based on both regression and classification models which brings novelty to this research. The best regressor and classifier model is selected to develop a web application that predicts and classifies the pIC50 values and biological activity classes of the drug molecules with the SMILES structure as the input.