A Generative Artificial Intelligence Copilot for Biomedical Nanoengineering

Area of Science:

• Biomedical Nanoscience
• Artificial Intelligence in Science
• Materials Science

Background:

• Large language models (LLMs) show promise for scientific research but often produce inaccurate or "hallucinated" outputs.
• Automated knowledge extraction from scientific literature is crucial for advancing research.

Purpose of the Study:

• To develop a generative AI tool, NanoSafari, for accurate knowledge extraction from biomedical nanoscience literature.
• To address scientific queries and provide reliable nanomaterial design parameters.

Main Methods:

• Developed the Grouped Iterative Validation based Information Extraction (GIVE) method.
• Extracted contextual nanoparticle characteristics from over 20,000 articles.
• Integrated an extracted database into a generative LLM.

Main Results:

• NanoSafari successfully extracted knowledge and addressed scientific queries.
• Blind evaluation by nanoscientists showed NanoSafari provided more reliable parameters than baseline models.
• Bench experiments validated the accuracy of the generated design parameters.

Conclusions:

• Generative AI tools like NanoSafari can automate learning from published scientific work.
• AI-based methods offer accurate and reliable references for biomaterial and bioengineering applications.
• NanoSafari demonstrates the utility of AI in overcoming LLM limitations for scientific research.