Benchmarking large language models for cell-free RNA diagnostic biomarker discovery

2026-06-11 · Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Artificial Intelligence & Machine Learning, Health & Medical Research, Life Sciences & Biology · Depth: Expert, short

Summary

A study benchmarked six large language models from OpenAI, Anthropic, and Google for diagnostic biomarker discovery using plasma cell-free RNA datasets. The evaluation spanned three clinical cohorts: Kawasaki disease versus multisystem inflammatory syndrome in children, active tuberculosis versus symptomatic respiratory controls, and myalgic encephalomyelitis/chronic fatigue syndrome versus sedentary controls. Models were assessed on literature-guided gene panel nomination for downstream machine learning and autonomous end-to-end classifier construction from raw count matrices. Despite prompt adherence issues, model-nominated panels effectively recapitulated canonical immune pathways and consistently outperformed random panels, even matching differential gene expression baselines in the tuberculosis cohort. End-to-end automation proved feasible but demonstrated model- and task-dependent performance, with one model approaching conventional levels for Kawasaki disease but decreasing for tuberculosis and ME/CFS.

Key takeaway

For Research Scientists developing diagnostic tools, this study indicates that large language models can significantly aid biomarker discovery. You should consider integrating LLMs for literature-guided gene panel nomination, as this approach consistently outperformed random selections and matched baselines in some contexts. While end-to-end automation is possible, carefully evaluate specific models and tasks, as performance varies. Prioritize LLM applications where robust panel identification is critical, and be prepared for task-dependent outcomes in fully automated classification pipelines.

Key insights

Large language models show promise for diagnostic biomarker discovery, particularly in gene panel nomination.

Principles

• Model-nominated panels recapitulate canonical immune pathways.
• End-to-end automation is feasible but model- and task-dependent.

Method

Benchmarking six LLMs on plasma cell-free RNA datasets across three clinical cohorts, evaluating literature-guided gene panel nomination and autonomous end-to-end classifier construction.

In practice

• LLMs can nominate diagnostic gene panels for machine learning.
• LLMs can construct end-to-end classifiers from raw count matrices.

Topics

• Large Language Models
• Biomarker Discovery
• Cell-free RNA
• Diagnostic Biomarkers
• Omics Data
• Machine Learning Benchmarking

Best for: AI Scientist, Research Scientist

Related on AIssential

• Knowledge Graphs and Reasoning LLMs for Finding Simple Yet Effective Transcriptomic Perturbation Predictors — Simple K-nearest neighbour models leveraging knowledge graphs, enhanced by RL-trained LLMs, offer competitive transcriptomic perturbation prediction.
• Active Evidence-Seeking and Diagnostic Reasoning in Large Language Models for Clinical Decision Support — LLMs exhibit a substantial performance gap in active, multi-turn clinical diagnosis compared to static, full-context evaluations.
• scLLM-DSC: LLM-Knowledge Enhanced Cross-Modal Deep Structural Clustering for Single-Cell RNA Sequencing — scLLM-DSC integrates LLM-derived biological semantics with transcriptomic features for superior scRNA-seq clustering.
• Benchmarking GPT Models for Conversational AI Systems: Can AI Read a Doctor’s Notes? — LLMs struggle with medical abbreviation expansion, especially context-dependent ones, despite larger models performing better.
• Automated reproducibility assessments in the social and behavioral sciences using large language models — Large language models can automate reproducibility assessments in social sciences, outperforming human reanalysts in qualitative agreement.
• Mamba-SSM with LLM Reasoning for Feature Selection: Faithfulness-Aware Biomarker Discovery — LLM reasoning significantly improves biomarker discovery performance by precisely removing confounders, even with imperfect biological recall.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.