Agentic AI integrated with scientific knowledge: laboratory validation in systems biology

Area of Science:

• Systems Biology
• Artificial Intelligence in Science
• Bioinformatics

Background:

• Automation and large language models (LLMs) accelerate scientific discovery but face challenges with logical reasoning.
• Integrating LLMs with laboratory automation requires robust frameworks for coherence and reliability.

Purpose of the Study:

• To develop a framework integrating LLM-based agents with laboratory automation for biological discovery.
• To improve the coherence and reliability of automated scientific workflows.
• To enable integrated hypothesis validation and refinement in systems biology.

Main Methods:

• A framework combining LLM agents with laboratory automation, guided by logical scaffolds.
• Symbolic relational learning, structured vocabularies, and experimental constraints.
• Automated cell-culture and metabolomics platforms for hypothesis validation.
• A graph database with controlled vocabularies for capturing hypotheses, experiments, and data.

Main Results:

• The integrated system demonstrated improved coherence and reliability in automated workflows.
• Novel interactions in Saccharomyces cerevisiae were identified, including glutamate-induced growth inhibition and aminoadipate's partial rescue of formic-acid stress.
• Extended existing ontologies and presented a novel representation of scientific hypotheses using description logics.

Conclusions:

• The developed framework shows potential for a reliable machine-driven discovery process in systems biology.
• Integration of AI agents and lab automation can overcome LLM limitations in logical structuring.
• This approach facilitates efficient hypothesis generation, validation, and refinement.