Enhanced in silico QSAR-based screening of butyrylcholinesterase inhibitors using multi-feature selection and machine learning

Butyrylcholinesterase inhibition offers one of the formulated solutions to tackle the aggravating symptoms of dementia that downgrades to cholinergic neuronal loss in Alzheimer's disease. We developed a QSAR model to facilitate the identification of effective butyrylcholinesterase inhibitors. The model employs multi-feature selection and feature learning, improving the in silico screening efficiency and accelerating drug discovery efforts. This study aims to integrate Human Intestinal Absorption (HIA) values of butyrylcholinesterase (BChE) target inhibitors and their 50% inhibitory concentration (IC50) with machine learning tools. The model was developed using chemical descriptors in combination with supervised machine learning classification algorithms. Random Forest Classifier algorithm proved to be the ultimate best fit for classification model metrics including log loss probability (0.04225), accuracy score (98.88%) and Matthew's correlation coefficient (0.98). Furthermore, a subset of the active dataset was used to study the regression based on HIA values using multi-feature selection and feature learning. The models were validated using precision, recall and F1 score for regression modelling. After integrating HIA data with existing machine learning algorithms, we observed a significant reduction of 89.63% in the number of inhibitors. The findings provide valuable pharmacological insights that can help in future design of drug development schemes different from conventional methods.