Evaluation and Prediction of Encapsulation of Bioactives in Cell-Based Microcarriers Using Machine Learning Approaches

Cell-based encapsulation systems can improve the stability and delivery of diverse bioactives, but predicting encapsulation efficiency is challenging due to various intrinsic and extrinsic factors. In the current study, a full factorial design was used to evaluate the influence of biochemical properties of yeast cells, chemical nature of bioactives, and ethanol level in the compound solution on the encapsulation efficiency of cell-based carriers. All compound-yeast-ethanol combinations showed linear trends between the loading yield and initial compound-to-cell ratio in the range chosen in the current study, implying that the encapsulation efficiency is constant regardless of initial compound-to-cell ratio, making it a good response variable. Higher overall encapsulation efficiency was achieved with a more hydrophobic compound and a lower ethanol level. Yeast cells with higher protein content achieved higher encapsulation efficiency for most of the compound-ethanol combination, except for retinol with 50% ethanol, where high-lipid cells exhibited higher efficiency. Overall, the chemical composition of the cell-based carriers has less significant effect on encapsulation efficiency compared to the other two factors. The most efficient predictive modeling pipeline for encapsulation efficiency consists of biochemical profiling of yeast cells with Fourier transform infrared spectroscopy (FTIR), feature extraction with a calibrated partial least square regression (PLSR) model, and prediction with a random forest model, obtaining a mean squared error (MSE) of 0.0095 and R² of 0.86 in the test set. Overall, the results highlight the potential of in silico pipeline and machine learning approaches to predict the encapsulation efficiency of cell-based carriers.