Natural language querying of biological databases with large language models.

Area of Science:

• Bioinformatics
• Computational Biology
• Artificial Intelligence in Life Sciences

Background:

• Biological knowledge bases are vast and complex.
• Accessing information often requires structured query languages.
• Natural language querying offers a more intuitive interface.

Purpose of the Study:

• To systematically assess current methods for natural language querying of biological data using large language models (LLMs).
• To identify optimal strategies for balancing accuracy and flexibility in LLM-based biological data mining.
• To propose requirements for effective benchmarks and future applications of LLMs in this domain.

Main Methods:

• Systematic assessment of natural language querying techniques.
• Evaluation of multiple large language model (LLM) agents.
• Analysis of agent interaction and human-in-the-loop feedback.

Main Results:

• Multiple interacting LLM agents achieve the best balance between accuracy and flexibility.
• Agent-to-agent challenges and human interaction enhance query performance.
• Current benchmarks are insufficient for robust assessment of natural language data mining systems.

Conclusions:

• Collaborative LLM agent systems offer a promising approach for natural language querying of biological knowledge bases.
• Development of standardized benchmarks is crucial for advancing the field.
• Future applications should focus on structured agent interactions and human oversight for optimal biological data mining.