OpenScientist: evaluating an open agentic AI co-scientist to accelerate biomedical discovery

Area of Science:

• Biomedical research
• Artificial intelligence in medicine
• Clinical data analysis

Background:

• Medical discovery relies on analyzing complex data, but human time and expertise are limited.
• Agentic artificial intelligence (AI) can automate scientific workflows, including data retrieval, analysis, and synthesis.
• Agentic AI has the potential to accelerate scientific discovery by overcoming human limitations.

Purpose of the Study:

• To introduce OpenScientist, an open-source agentic AI co-scientist.
• To accelerate biomedical discovery through semi-autonomous investigation of scientist-defined queries.
• To generate clinically relevant and verifiable scientific insights.

Main Methods:

• OpenScientist was evaluated by domain experts in four clinical case studies.
• Case studies included Alzheimer's disease biomarker analysis, plasma proteomic survival prediction, and neurofibrillary tangle data investigation.
• Hypothesis generation and validation in a multiple myeloma dataset with a randomized negative control were also performed.

Main Results:

• OpenScientist completed analyses in minutes that typically require weeks to months of human effort.
• It identified %ptau217 as a key predictor of amyloid PET status.
• The AI generated a plasma proteomic survival model, proposed a mechanism for tau pathology, and validated multiple myeloma hypotheses.

Conclusions:

• Open-source, auditable agentic AI can significantly support clinical research.
• OpenScientist demonstrates the capability to generate hypotheses and discover insights from complex datasets.
• Agentic AI can accelerate the pace of biomedical discovery and clinical research.