Evaluation of Machine Learning Methods for Identifying Carbonic Anhydrase-II Inhibitors as Drug Candidates for Glaucoma

Glaucoma is a leading cause of irreversible blindness, primarily managed by lowering intraocular pressure (IOP). Carbonic Anhydrase-II (CA-II) inhibitors play a crucial role in this treatment by reducing aqueous humor production. However, existing CA-II inhibitors often suffer from poor selectivity, side effects, and limited bioavailability, highlighting the need for more efficient and targeted drug discovery approaches. This study uses machine learning-driven Quantitative Structure-Activity Relationship (QSAR) modeling to predict CA-II inhibition based on molecular descriptors, significantly enhancing screening efficiency over traditional experimental methods. By evaluating multiple machine learning models, including Support Vector Machine, Gradient Boosting, and Random Forest, we identify SVM as the most effective classifier, achieving the highest accuracy (83.70%) and F1-score (89.36%). Class imbalance remains challenging despite high sensitivity, necessitating further improvements through resampling and hyperparameter optimization. Our findings underscore the potential of machine learning-based virtual screening in accelerating CA-II inhibitor identification and advocate for integrating AI-driven approaches with traditional drug discovery techniques. Future directions include deep learning enhancements and hybrid machine learning-docking frameworks to improve prediction accuracy and facilitate the development of more potent and selective glaucoma treatments.