Predicting and optimizing viscosity of dental resin composites with Gaussian process regression and Bayesian...

Area of Science:

• Materials Science
• Data Science
• Biomaterials Engineering

Background:

• Dental resin composites require specific handling properties, such as optimal viscosity, for clinical application.
• Traditional methods for optimizing composite formulations are often time-consuming and may not explore the full formulation space efficiently.

Purpose of the Study:

• To develop and validate a machine learning framework utilizing Gaussian Process Regression (GPR) and Bayesian Optimization (BO).
• To predict and optimize the viscosity of resin composites at two distinct shear rates (0.0106 s⁻¹ and 74.4 s⁻¹).

Main Methods:

• Fifty-four resin composite formulations were prepared with varying filler compositions (two main fillers, fumed silica).
• Viscosity was measured using rheometry; log-transformed values were used as targets for GPR models trained with 10-fold cross-validation.
• Bayesian Optimization (BO) with Probability of Improvement was employed to identify optimal formulations from a large candidate pool (67,140).
• Shapley Additive Explanations (SHAP) analysis was used for feature interpretation.

Main Results:

• GPR models showed significant correlation between predicted and experimental viscosity values (p < 0.001).
• SHAP analysis identified key formulation parameters influencing viscosity: fumed silica content at low shear rates, and main filler particle size/surface treatment at high shear rates.
• BO successfully optimized viscosity, achieving target values within seven iterations, demonstrating efficient navigation of the formulation space.

Conclusions:

• Gaussian Process Regression (GPR) and Bayesian Optimization (BO) offer a powerful data-driven approach for designing dental resin composites.
• This framework enables rational optimization of rheological properties, crucial for clinical handling characteristics like sculptability and extrudability.
• Future work should involve expanding datasets and incorporating multi-objective optimization to balance viscosity with other critical material properties.