Beyond Trial and Error: Accelerated Bayesian Design of Dental Adhesives

The development of new dental adhesive formulations traditionally requires extensive laboratory testing due to the complex interplay of components and multiple performance criteria that must be optimized. Here, we present an innovative approach combining active learning and Bayesian optimization (BO) that accelerates the development of high-performing dental adhesives. Leveraging 91 historical trial-and-error experiments involving 43 components, we identified 9 key components critical for adhesive performance. This focused design space enabled our machine learning framework to efficiently converge on 4 novel formulations with high-performing shear bond to dentin within 40 experiments. Our optimization strategy revealed critical insights into formulation design, particularly the crucial balance between organic solvents and functional monomers that governs adhesive performance. This approach not only achieved a 19% increase in the modal distributions compared to traditional formulations but also significantly reduced material consumption and laboratory resources through rapid convergence to optimal solutions. The success of this methodology demonstrates its potential for efficient dental materials development, offering a pathway to more sustainable and cost-effective innovation in dental adhesive technology. This framework can be readily adapted to optimize other dental materials, potentially transforming how we approach formulation development in dental materials research.