Enhanced formulation of precision probiotics through active machine learning

Abstract The human gut microbiome is crucial to health, with dysbiosis increasingly linked to disease. Precision probiotics offer a promising approach to restoring microbial balance, but ensuring probiotic viability through gastrointestinal transit remains a challenge. This study applies an advanced active machine learning (ML) approach to predict how excipients affect the growth of Lactobacillus plantarum, a commonly used probiotic. State-of-the-art experiments were carried out to complement the ML study. Starting with five known excipient-probiotic interactions, we apply active ML over three rounds to predict the effects of 116 excipients, iteratively refining model certainty and accuracy. Five ML models—neural networks, gradient boosting, logistic regression, random forest, and support vector machines—were trained and evaluated, with the final model achieving certainty levels close to 90%. Unlike previous methods, which retrained new models per iteration, our approach continuously optimized a single model, enhancing prediction stability and reducing uncertainty spread. These results highlight the potential of active ML to support accurate excipient selection in probiotic formulations.