Performance Comparison of Machine Learning Methods in Discovery of BACE-1 Inhibitors in Alzheimer's Disease Therapy

Abstract

Alzheimer's disease (AD) poses a significant challenge in the realm of neurodegenerative disorders, necessitating effective therapeutic interventions. One promising approach involves the discovery of ?-secretase 1 (BACE-1) inhibitors, pivotal in mitigating amyloid-? peptide accumulation, a hallmark of AD pathology. In this study, we compare the performance of three prominent machine learning methods, namely Gradient Boosting Machine (GBM), Random Forest (RF), and Support Vector Machine (SVM) in the discovery of BACE-1 inhibitors. Leveraging the BACE dataset sourced from MoleculeNet, comprising quantitative and qualitative binding results of compounds, we explored the classification efficacy of these methods. Our experimental results reveal distinct precision, recall, and accuracy metrics for each method, showcasing RF with precision and accuracy scores of 1.00 and 99.67%, respectively, followed by GBM and SVM. Furthermore, feature importance analysis underscores pIC50 as the most influential attribute across all methods, emphasizing its pivotal role in classifying BACE-1 inhibitors. Additionally, RF prioritizes Estate as the second most important feature, while AlogP emerges as GBM's secondary significant attribute. These findings shed light on the efficacy of machine learning techniques in identifying potential therapeutics for AD, offering insights into feature importance variations among methods and highlighting the significance of diverse molecular descriptors in drug discovery.