Optimization Of Whitening Toothpaste Using Artificial Intelligence-Based Bayesian Optimization

To optimize the whitening efficacy and stability of low-concentration (0.01–2.5%) hydrogen peroxide (HP) toothpaste formulations and to assess the feasibility of applying AI-based Bayesian optimization in dental formulation design. A Bayesian optimization framework was applied, using Gaussian Process Regression to design the acquisition functions, specifically Upper Confidence Bound and Expected Hypervolume Improvement. The formulation space included stabilizers such as tripolyphosphate and polyvinylpyrrolidone. Whitening efficacy was evaluated using an in vitro hydroxyapatite (HAP) staining model: (1) staining HAP discs, (2) pre-treatment color measurement (Spectro2guide 45°, BYK, Germany), (3) toothpaste application in artificial saliva (1:3) for 2 min at 37°C, and (4) post-treatment color measurement (ΔE = Ebefore - Eafter). Bayesian Optimization facilitated the identification of formulations achieving significant whitening efficacy at HP concentrations as low as 0.01%, which was generally unattainable. Partial Dependence Analysis using a machine learning model revealed critical formulation parameters. Compared to conventional approaches that individually vary the concentration of each compound, AI-driven optimization efficiently explored the formulation space, enabling the discovery of novel compositions. This study demonstrates that AI-driven optimization enables the achievement of effective whitening at low HP concentrations, which may benefit individuals concerned about peroxide exposure. The ability to enhance whitening efficacy through AI-based formulation design provides new insights for dental materials research.