AI Data Scientist at MUST AI-X Enables Autonomous Analysis of Massive Data: “OpenClaw” Emerges in Biomedicine
AI Data Scientist at MUST AI-X Enables Autonomous Analysis of Massive Data: “OpenClaw” Emerges in Biomedicine
Recently, a research team led by Professor Zhang Kang, Dean of the Artificial Intelligence Cross Disciplinary Research Institute (AI-X) at Macau University of Science and Technology, in collaboration with Professor Zhao Yi’s team from the Institute of Computing Technology, Chinese Academy of Sciences, as well as multiple research and clinical institutions, published a research article in the international journal Nature Biomedical Engineering, titled “A multi-agent LLM framework with self-evolving capabilities for autonomous, tool-aware biomedical data analyses.”
This study proposes and systematically validates a novel biomedical artificial intelligence framework—BioMedAgent. Designed to address challenges in biomedical research, including complex analytical workflows, high technical barriers, and insufficient automation, the framework explores the development of an “AI data scientist” capable of autonomously planning, invoking tools, and completing multi-step analytical tasks, thereby providing a new intelligent solution for biomedical data analysis.
As one of the representative achievements of the “AI+X” interdisciplinary initiative of the Artificial Intelligence Cross Disciplinary Research Institute at Macau University of Science and Technology, this study reflects the deep integration of artificial intelligence with medical and health sciences, life sciences, and data science. It also demonstrates the Institute’s efforts to advance artificial intelligence toward real-world application scenarios centered on complex problems.
With the rapid development of artificial intelligence technologies, significant potential has been demonstrated in areas such as medical imaging analysis, electronic health record interpretation, and multi-omics research. However, in real-world and complex research scenarios involving multi-step reasoning, tool invocation, and cross-modal data integration, enabling artificial intelligence to complete the full process—from task understanding and analytical planning to execution and summarization—like an experienced researcher remains a major challenge.
To address these challenges, the research team developed BioMedAgent, a biomedical data analysis framework based on a multi-agent large language model. The system integrates multiple functionally coordinated modules, linking task planning, code generation, tool invocation, result execution, and feedback optimization into a complete analytical loop of “thinking–planning–execution–reflection.” Compared with traditional models, BioMedAgent can automatically decompose complex tasks based on natural language instructions and invoke bioinformatics tools and database interfaces to construct relatively complete data analysis workflows, thereby enhancing the automation of complex scientific tasks.
In addition, the study introduces mechanisms of Interactive Exploration (IE) and Memory Retrieval (MR), which support experience accumulation and strategy optimization during task execution. Professor Zhao Yi noted that BioMedAgent is not limited to fragmented instruction execution or auxiliary programming, but instead establishes a complete analytical pipeline encompassing “thinking–planning–execution–reflection.” A key feature of the system is its ability to accumulate experience during task execution and error correction through interactive exploration and memory retrieval, continuously optimizing analytical strategies and code invocation methods, and thereby progressively improving its capability to address complex biomedical problems.
The results show that BioMedAgent achieved strong performance in benchmark tests, with an overall task success rate of 77%, including 94% for omics analysis tasks and 90% for machine learning tasks. Compared with existing mainstream large language model agent approaches, BioMedAgent demonstrated clear advantages across multiple task categories and exhibited strong generalization capability in external independent benchmarks.
Beyond benchmark testing, the research team further validated the application potential of BioMedAgent in multiple real-world research scenarios. In cross-omics studies, the system can automatically perform integrated analysis of RNA-seq and single-cell RNA sequencing data based on natural language instructions, identify differentially expressed genes, and determine their cellular origins, with results consistent with existing research findings. In machine learning modeling tasks, BioMedAgent can complete the entire process from model construction and training to evaluation without manual programming intervention, successfully reproducing key findings from previous studies. In pathological image analysis tasks, the system further improved multi-class cell segmentation accuracy by automatically integrating resolution enhancement and cell segmentation algorithms, demonstrating strong practical value.
The research team emphasized that BioMedAgent is not intended to replace researchers, but rather to serve as an intelligent assistant for scientific research and clinical applications, helping researchers improve efficiency and reduce technical barriers when dealing with complex data and multi-step analytical tasks. This work highlights the potential of artificial intelligence to evolve from single-tool assistance to higher-level autonomous collaborative analysis, and provides a new technological pathway for multi-tool integration, multi-step reasoning, and the automation of complex research workflows.
Professor Zhang Kang stated that the Artificial Intelligence Cross Disciplinary Research Institute at Macau University of Science and Technology will continue to promote the deep integration of artificial intelligence with medical and health sciences, life sciences, engineering, and data science. Focusing on complex real-world problems, the Institute will carry out collaborative research and explore a new paradigm of “autonomous driving” human–AI collaborative research, providing strong support for life science research, drug discovery, and clinical translation.
Link：https://www.nature.com/articles/s41551-026-01634-6
