MR-MAS: An End-to-End Multi-Agent System for Autonomous Mendelian Randomization Analysis.

MR-MAS, an automated multi-agent system, achieves stable and reproducible Mendelian Randomization analyses with consistent causal estimates across multiple runs and models.

• Why it matters: Efficient, standardized, and reproducible MR workflows are limited by fragmented processes and reliance on specialized expertise, hindering biomedical research progress.
• What they did: The system integrates data acquisition, causal effect estimation, interpretation, and manuscript preparation using large language models and external tools, supported by a structured MR Knowledge Base, tested across diverse scenarios.
• The result: MR-MAS demonstrated stable execution, reproduced literature-supported causal signals, and showed improved code-repair performance, suggesting it can facilitate scalable and reproducible MR analyses in biomedical research.