Data-Driven Design: Leveraging Predictive Modeling for the Development of Soluble and Insoluble Glasses for Medicine

Abstract

The rational design of biomaterials that interact with biological systems to promote healing and regeneration remains a central challenge in medicine. Traditional materials discovery approaches, particularly for bioactive glasses, are limited by trial-and-error methods. This thesis adopts a data-driven approach aligned with the Materials Genome Initiative (MGI) to predict composition-property relationships in both soluble and insoluble glasses for medicine. A systematic investigation of 23 multi-component soluble borate glasses enabled the development of statistical models capable of predicting the individual and interaction effects of glass constituents on structure, dissolution, ion release, and cytocompatibility. In parallel, 16 aluminosilicate glass microsphere formulations were developed for transarterial radioembolization (TARE), optimized for thermal and chemical stability, low cytotoxicity, and compatibility with CT and SPECT imaging. Across both systems, predictive modeling enabled the identification of optimal formulations, demonstrating the power of integrated experimental and computational strategies in advancing bioactive glass design for clinical applications.