Ensemble Learning in QSAR: Unveiling Insights into PTP1B Inhibitors for Diabetes Mellitus Treatment

Diabetes Mellitus (DM) stands as a metabolic disorder characterized by abnormal elevations in blood glucose levels. It encompasses various forms, including type 1, type 2, maturity-onset diabetes of the young (MODY), gestational diabetes, neonatal diabetes, and secondary instances arising from endocrinopathies, steroid usage, and other contributing factors. In accordance with data from the International Diabetes Federation (IDF), the prevalence of diabetes in Indonesia reached 4.8 percent in 2012. To address this public health challenge, a Quantitative Structure-Activity Relationship (QSAR) model employing the Ensemble Method was developed for diabetes drug discovery. QSAR represents a computational or mathematical modeling approach employing chemometrics, aimed at establishing statistically significant correlations between functionality and molecular structure. Within this research, the Ensemble Method was deployed, specifically examining the comparative performance of Random Forest, AdaBoost, and XGBoost algorithms. This ensemble approach effectively predicted the activity of the PTP1B inhibitor. Notably, it involved comprehensive feature selection and parameter tuning, ultimately identifying XGBoost as the superior algorithm, employing 90% of the selected features. The performance metrics of XGBoost were notable, with an accuracy rate of 94.84%, a recall rate of 94.74%, a precision rate of 94.86%, and an F1-score of 95.27%.