Stanford researchers unveil Paper2Agent system to convert studies
- Scientific literature is shifting from static text to active infrastructure through a new system called Paper2Agent, which packages published studies, code, methods, and raw data into automated tool...
- Stanford University computer scientist James Zou and his colleagues tested the system on 100 computational biology papers.
- While researchers routinely share programming code on platforms like GitHub, utilizing another scientist's software often requires substantial manual setup.
Scientific literature is shifting from static text to active infrastructure through a new system called Paper2Agent, which packages published studies, code, methods, and raw data into automated tool sets for artificial intelligence assistants. Researchers published details of the system on September 16 in Nature, describing it as a virtual author capable of working with new data around the clock.
Converting Computational Biology Papers Into Active Tool Sets
Stanford University computer scientist James Zou and his colleagues tested the system on 100 computational biology papers. The automated conversion succeeded with 74 of them, while the remaining studies faced technical obstacles such as missing code, unavailable data, or software that failed to execute. For the successfully converted papers, the resulting Paper2Agent tool sets ran analyses of the published work and answered specific questions about the results.
Streamlining Code Reuse Across Research Teams
While researchers routinely share programming code on platforms like GitHub, utilizing another scientist’s software often requires substantial manual setup. Peter Clark, a senior research director at the Allen Institute for Artificial Intelligence in Seattle who was not involved in developing the system, noted that investigators typically spend hours troubleshooting how to make external code run. General-purpose assistants like Claude can generate replacement code on the fly, but Paper2Agent differs by bundling established code into reusable tools, verifying them against reference outputs, and writing step-by-step instructions for AI assistants. Clark described the framework as an exciting step toward research papers that actively assist scientists.
Fostering Potential Collaborations Through Virtual Authors
Beyond helping individual researchers execute complex methods, the paper-agents are designed to discover related work and connect authors. Zou reported that his team has generated over 100 different paper-agents from their own publications, and these tools have already helped identify peers for prospective research collaborations. Clark cautioned that the system focuses tightly on code and methodology reuse rather than open-ended conversational queries, noting that corresponding authors remain essential for addressing the unwritten nuances behind a study.