Benchmarking large language models for cell-free RNA diagnostic biomarker discovery
AI language models show promise for discovering disease biomarkers in blood, matching traditional methods while revealing where computational limits lie.
Cornell benchmarked six frontier LLMs from OpenAI, Anthropic, and Google on cfRNA biomarker discovery tasks across three disease cohorts, showing LLMs can match differential expression baselines for infectious disease classification while revealing model/task-specific limitations. This defines a practical capability envelope for AI-assisted liquid biopsy biomarker discovery.
What the study was
- Study design
- Benchmarking study (6 LLMs vs 3 cfRNA clinical cohorts)
- Population
- Multi-disease plasma cfRNA datasets (Kawasaki/MIS-C, TB, ME/CFS)
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- Nature Communications
Why it surfaced
Nat Commun; first rigorous head-to-head LLM benchmark on cfRNA biomarker discovery; directly informs AI integration in liquid biopsy platforms.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.