A ReAct Agentic AI System for Natural Language Querying and Statistical Analysis of The Cancer Genome Atlas Clinical Data

A ReAct agentic AI system achieves 93.4% accuracy in natural language querying and analysis of TCGA clinical data, surpassing rule-based and LLM baselines.

• Why it matters: Accessing complex, heterogeneous TCGA data is challenging, limiting efficient clinical and research insights; an intelligent system can bridge this gap.
• What they did: The system employs a large language model as an autonomous ReAct agent, utilizing eight computational tools to perform data extraction, statistical analysis, and validation across 440 benchmark queries.
• The result: It enables precise, auditable analysis of clinical data, especially in fields lacking curated information, demonstrating the importance of tool design and reasoning loops over model size alone.