Prediction of the insulin growth factor release capacity from polymeric carriers on additively manufactured Ti-6Al-4V scaffolds using machine learning

Abstract

Electron beam powder bed fusion (EB-PBF) has produced Ti-6Al-4V scaffolds that show promise for bone engineering. The current study employed machine learning (ML) analyses on 60 samples of experimental polymeric carriers that contained insulin growth factor-1 (IGF-1) on Ti-6Al-4V scaffolds and applied their porous structure for enhanced drug transport. Important determinants were pore size, drug release mass, time, and carrier ratio. We systematically developed and evaluated two ensemble regression algorithms, the Gradient Boosting (GB) Regressor and the Random Forest (RF), using several statistical metrics, such as the coefficient of determination (R²), the root mean squared error (RMSE), the mean squared error (MSE), and the mean absolute error (MAE). Bayesian optimization methods were used to improve model generalization and robustness by optimizing hyperparameters. Among the models used, RF performed the best with an R² value of 0.87, signifying an excellent degree of agreement between predicted and experimental results and demonstrating symmetric error distribution for high reliability and resistance towards overfitting. This association between the scaffolds and machine learning will further advance drug stability and efficacy as well as controlled drug delivery technologies while making biomaterial development more efficient and data-driven.