Harnessing Data-Intelligence-Intensive Multi-Agent System for Life Science Research

Advancements in high-throughput sequencing technologies and artificial intelligence offer unprecedented opportunities for groundbreaking discoveries while posing significant analytical challenges. This study introduces a data-intelligence-intensive scientific research paradigm that synergizes human expertise with AI to facilitate hypothesis-free exploratory research in life science. We propose a multi-agent system (DII-MAS) based on large language models (LLMs), enabling efficient human-agent interaction, agent group management, interdisciplinary knowledge empowerment, and continuous learning. This novel framework is demonstrated through the construction of a human lung cell atlas, showcasing its capability to overcome the limitations of standalone AI applications, improve research efficiency, and adapt to complex life science tasks. This study substantiates three key hypotheses: the collective intelligence workflow can significantly propel life science tasks, proactive interactions within DII-MAS mitigate comprehension biases and incomplete information issues, and continuous learning empowers DII-MAS to make optimal decisions and tool selections. The contributions of this study comprise the delineation of a data-intelligence-intensive research paradigm, the development of DII-MAS, and the introduction of novel evaluation metrics for agent performance. This study underscores the potential of integrating AI with expert knowledge to accelerate discoveries and navigate uncharted territories in life sciences. ### Competing Interest Statement The authors have declared no competing interest.