Experimental and machine learning analysis of structure-property relationships in GPTMS-crosslinked gelatin scaffolds for bone regeneration.

This study develops bioactive freeze-dried porcine gelatin scaffolds crosslinked with glycidoxypropyl trimethoxysilane (GPTMS) and evaluates their suitability for bone-regeneration applications through integrated experimental characterization and mechanism-aware machine learning. Gelatin/GPTMS scaffolds containing 0.9%, 1.2%, and 1.5% GPTMS exhibited interconnected porous architectures, predominantly amorphous hybrid networks, and successful siloxane crosslink formation confirmed by SEM, XRD, and FTIR analyses. The 0.9% GPTMS formulation provided the most favorable mechanical balance, with the highest compressive strength and elastic modulus, whereas higher GPTMS contents increased pore size and accelerated degradation; importantly, GPTMS-containing scaffolds showed apatite-forming ability in simulated body fluid, confirming mineralization capacity absent in pure gelatin. A physics-guided Gaussian Process Regression framework further predicted compressive strength, elastic modulus, and degradation behavior with favorable accuracy within the available dataset, while SHAP-based interpretation identified porosity and GPTMS-derived FTIR features as the dominant mechanistic drivers. Overall, the results demonstrate that GPTMS-crosslinked gelatin scaffolds provide a tunable combination of porosity, mechanical competence, biodegradability, and bioactivity, while mechanism-aware machine learning offers a data-efficient route for scaffold-property prediction and biomaterials design.