Why Read a Research Paper When You Can Turn It Into an AI Agent?
Stanford computer scientist James Zou and colleagues described an open-source framework called Paper2Agent in Nature on 16 September. The tool takes a paper along with its accompanying codebase, data or other supplementary material and automatically extracts core workflows, then builds a tested, runnable toolkit that can be used on other datasets. The researchers tested it across statistics, econometrics and astrophysics, focusing proof-of-concept demonstrations on computational biology.
The team first used AlphaGenome, a deep-learning model that predicts how DNA mutations affect gene regulation. Given the corresponding documentation and code, Paper2Agent produced 22 tools covering different aspects of AlphaGenome functionality in about 45 minutes with no human intervention, on a personal laptop and for less than US $15 in computing costs. A testing agent validated all 22 tools against reference results, diagnosing failures and attempting fixes up to six times per function before potentially dropping a tool. The validated tools were packaged into a Model Context Protocol server and connected to Claude Code. According to the researchers' analysis, the resulting agent outperformed standard Claude given the AlphaGenome codebase and a specialist AI co-scientist tool called Biomni.
The researchers then converted two additional papers into agents and linked them with the AlphaGenome agent. Prompted to investigate the genetic basis of psoriasis, the three agents identified the little-understood gene GPR137 as a likely causal factor and proposed 10 validation approaches. A human researcher selected one; the resulting analysis found that silencing GPR137 produced gene activity changes strikingly similar to those caused by the psoriasis-linked variant in immune cells.
Of 100 computational biology papers they tried, 26 failed to convert into agents, often because of incomplete code, missing documentation or software packages that could not be made to work. Zou said this can expose missing information, code errors or discrepancies between a paper and its implementation. Olivier Elemento of Weill Cornell Medicine, who peer-reviewed the study, called it "a real advance in terms of how we think about the publication process." Dongping Chen of the University of Maryland said making papers executable through an agentic interface is "quite compelling." Artur Skowroński of VirtusLab noted in a blog post that Paper2Agent could help bring papers to life in classrooms. One day after the Nature paper, the team unveiled Virtual Biotech, described in Science, and posted a Paper2Agent-generated version of that paper. The Paper2Agent manuscript was also fed into Paper2Agent, creating an agent at paper2agent.ai.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
Have you ever read a paper in Science or Nature and thought, “Man, that research was so cool. I wish I could try that method on my own data,” only to spend a week wrestling with someone else’s undocumented repo, broken dependencies, and half-finished readme.txt? Well, now you can, more or less. Say hello to Paper2Agent, a new open-source framework that transforms academic reports into interactive AI agents you can talk to. Give it a paper, along with the accompanying codebase, data, or other supplementary material,