Interpretable AI in Tissue Engineering: XGBoost and SHAP for PLGA Scaffold Biocompatibility

Abstract

The refinement of scaffold materials is essential in tissue engineering to promote cellular growth and tissue regeneration. This study applied Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) to predict and interpret the biocompatibility of poly(lactic-co-glycolic acid) (PLGA)-based scaffolds. A dataset of 10,010 synthetic samples was analyzed, examining key scaffold features such as Young's Modulus, Ultimate Tensile Strength, Strain at Failure, Compressive Modulus, Pore Size, Porosity, and Degradation Time. The XGBoost model demonstrated high predictive accuracy with a Root Mean Square Error (RMSE) of 2.59. SHAP analysis identified Young's Modulus, Strain at Failure, and Ultimate Tensile Strength as the most influential factors affecting biocompatibility. These findings highlight the critical role of mechanical properties in scaffold performance, particularly in cell adhesion and tissue integration. This research offers a data-driven framework for optimizing scaffold designs and integrating machine learning predictions with biological insights for improved tissue engineering applications.