QSAR Modeling of Beta-2 Adrenergic Receptor Ligands Using Molecular Descriptor–Based Machine Learning

Abstract

The Beta-2 Adrenergic Receptor (ADRB2) is a well-characterized G protein–coupled receptor widely studied in pharmacology and drug discovery. In this study, quantitative structure–activity relationship (QSAR) models were developed using molecular descriptor–based machine learning approaches to predict the activity of ADRB2 ligands. A curated dataset of 745 compounds with experimentally determined IC₅₀ values was obtained from the ChEMBL database. Two-dimensional molecular descriptors were calculated and preprocessed to remove low-variance and highly correlated features, resulting in a refined feature set for model development. The dataset was categorized into active and inactive compounds and divided into training and testing subsets. Four machine learning algorithms. Logistic Regression, Support Vector Machine, Gradient Boosting, and Random Forest were implemented and evaluated using accuracy, precision, recall, F1-score, and ROC-AUC metrics. Among the models, Random Forest achieved the best performance, with an accuracy of 89.26%, F1-score of 89.87%, and AUC of 0.926, followed by Gradient Boosting with an accuracy of 87.92% and AUC of 0.922. Analysis of physicochemical descriptors indicated that hydrogen-bond donor capacity (nHD) shows a statistically significant association with variations in compound activity toward ADRB2, while lipophilicity (LogP) and hydrogen-bond acceptor count (nHA) do not exhibit statistically significant differences between activity classes. Overall, the results demonstrate that molecular descriptor–based machine learning models, particularly ensemble methods, provide an effective framework for predicting ADRB2-related compound activity and support the prioritization of candidate molecules in computational drug discovery.