BacterAI maps microbial metabolism without prior knowledge.

Area of Science:

• Microbiology
• Artificial Intelligence
• Metabolic Engineering

Background:

• Autonomous experimentation can significantly increase microbiology research throughput.
• A major limitation is the scarcity of large datasets required for training artificial intelligence (AI) systems.
• Existing methods often require prior knowledge or extensive data, hindering the study of novel or understudied organisms.

Purpose of the Study:

• To introduce BacterAI, an automated science platform for mapping microbial metabolism.
• To demonstrate an AI system that requires no prior knowledge for organism study.
• To enable unbiased, autonomous investigation of microbial metabolism, particularly for organisms with limited existing data.

Main Methods:

• BacterAI converts scientific questions into games played with laboratory robots.
• The AI agent distills experimental findings into interpretable logical rules.
• Transfer learning is employed to accelerate investigations in new environments or with complex media.

Main Results:

• BacterAI successfully mapped the amino acid requirements for Streptococcus gordonii and Streptococcus sanguinis.
• Transfer learning demonstrated accelerated learning for new conditions and media compositions (up to 39 ingredients).
• The platform proved effective for organisms lacking pre-existing training data.

Conclusions:

• BacterAI facilitates the autonomous and unbiased study of microbial metabolism.
• The scientific gameplay approach overcomes data limitations in AI-driven biological research.
• This platform has the potential to significantly advance microbiology by enabling the study of previously intractable organisms.