Enhanced formulation of precision probiotics through active machine learning.

Area of Science:

• Microbiology
• Bioinformatics
• Pharmacology

Background:

• The human gut microbiome plays a vital role in overall health.
• Microbial imbalances (dysbiosis) are linked to various diseases.
• Probiotic interventions aim to restore microbial balance, but ensuring survival during gastrointestinal transit is difficult.

Purpose of the Study:

• To employ an advanced active machine learning (ML) approach to predict the impact of excipients on the growth of *Lactobacillus plantarum*.
• To enhance the selection of excipients for improved probiotic formulation stability and efficacy.

Main Methods:

• Utilized an active ML strategy over three iterative rounds to predict excipient effects on *Lactobacillus plantarum* growth.
• Trained and evaluated five ML models: neural networks, gradient boosting, logistic regression, random forest, and support vector machines.
• Compared continuous model optimization against retraining new models per iteration.

Main Results:

• Successfully predicted the effects of 116 excipients on probiotic growth, achieving model certainty levels near 90%.
• The active ML approach demonstrated enhanced prediction stability and reduced uncertainty spread compared to traditional methods.
• Validated ML predictions with complementary state-of-the-art experimental data.

Conclusions:

• Active ML is a powerful tool for predicting excipient-probiotic interactions.
• This method can significantly improve the accuracy and efficiency of selecting excipients for probiotic formulations.
• The findings support the development of more effective precision probiotic therapies.