Encapsulation of Lactobacillus reuteri and Its application in maras-type ice cream using data optimized by artificial neural networks

Abstract In the first stage of this study, the encapsulation efficiency (EE) of Lactobacillus microorganisms was predicted using an artificial neural network (ANN) model. A total of 144 datasets obtained from the literature were used for model development. The ANN was trained using a multilayer structure and the Levenberg–Marquardt (trainlm) backpropagation algorithm, while a genetic algorithm (GA) was applied to optimize model performance. In the model, the Lactobacillus strain, coating material, and encapsulation method were defined as input variables, whereas EE, process stress, and storage stability were considered as output variables. The dataset was divided into training (70%), validation (15%), and testing (15%) subsets. The validation coefficient of the model was determined as 0.90066, indicating high predictive reliability. In the second stage, optimal conditions obtained from the ANN model were identified as Lactobacillus reuteri , 2.5% alginate, and the emulsion method. Under these conditions, the encapsulation efficiency reached 91.44%. To assess process resistance, both non-encapsulated (L) and encapsulated (E) probiotics were incorporated into ice cream. The post-processing viability was 86.5% for L and 96.32% for E, showing a statistically significant difference ( p <.05). The ice cream samples were stored for 90 days, and probiotic viability was monitored throughout the storage period. At the end of storage, probiotic levels remained above 10⁶ CFU/g in all samples, although significant differences between E and L groups were observed ( p <.05). Additionally, probiotic incorporation did not adversely affect sensory properties, and no significant difference in overall acceptability was found ( p >.05). Graphical Abstract