Data-driven modeling of dose-dependent chlorhexidine release from glass ionomer cements using gradient boosting models

Abstract

Objectives: Long term prediction of antimicrobial release from bioactive dental materials remains challenging due to complex, nonlinear release kinetics. This study presents the application of the extreme gradient boosting (XGBoost) predictor to model sustained chlorhexidine (CHX) release from glass ionomer cements (GICs), enabling accurate long-term predictions. Methods: XGBoost models were trained using a comprehensive dataset of cumulative CHX release from CHXhexametaphosphate (CHX-HMP) functionalized GICs measured over 663 days across 1%, 2%, 5% and 10% dose levels. Cross-validated models demonstrated accurate prediction of the CHX release within the 663-day observation period. The model performance has been maximized using Bayesian hyperparameter optimization. Closed-form predictive equations have been developed for all doses to forecast the CHX release beyond 663 days. An online graphical user interface has been developed on the Streamlit platform. Results: The optimized XGBoost models demonstrated high predictive accuracy for cumulative CHX release and daily release rates across all dose levels. Exponential forecasting equations achieved R2 scores greater than 0.92 for all dose levels. The conformal prediction technique provided reliable prediction intervals for the daily CHX release rate. Significance: This study presents accessible, and uncertainty-aware models for predicting long term CHX release from antimicrobial GICs. The approach enables reliable forecasting beyond the experimental time window, supporting the design and clinical application of bioactive dental materials. The online tool facilitates practical adoption by researchers and clinicians, addressing the challenge of long-term antimicrobial release prediction in restorative dentistry.