Machine Learning Reveals Quantitative Amino Acid Preferences in Bifidobacterium longum Growth

Area of Science:

• Microbiology
• Computational Biology
• Nutritional Science

Background:

• Bifidobacterium longum is a key human gut bacterium with well-understood carbohydrate metabolism.
• The precise role of amino acids in supporting B. longum growth is not fully quantified.
• Understanding amino acid requirements is crucial for optimizing probiotic formulations and gut health research.

Purpose of the Study:

• To quantitatively determine the amino acid preferences and requirements of Bifidobacterium longum subsp. longum JCM 1217^T.
• To develop a machine-learning framework for designing optimized, reduced-complexity growth media.
• To identify specific amino acids that significantly impact B. longum growth dynamics.

Main Methods:

• Genome-based pathway analysis to predict nutritional requirements.
• Growth phenotyping in chemically defined media under various conditions.
• Iterative machine-learning algorithms (regression models, genetic algorithms, SHAP analysis) for medium optimization and key factor identification.

Main Results:

• Cysteine was confirmed as an auxotrophy, but complete amino acid mixtures supported superior growth.
• Optimized media formulations reduced total amino acid input by up to 77.2% while maintaining comparable growth.
• SHAP analysis identified tyrosine as critical for maximum cell density and glutamate, leucine, and valine for reduced lag time.

Conclusions:

• B. longum's amino acid requirements are more complex than simple auxotrophy, with specific amino acids influencing distinct growth parameters.
• A machine-learning approach enables the design of streamlined, defined media for bacterial cultivation.
• This framework facilitates precise nutritional studies and the development of targeted probiotic strategies.