Paper2Agent turns 74 of 100 biology papers into AI agents

Of 100 computational biology papers put through a new framework called Paper2Agent, 74 were successfully converted into working AI agents that answered benchmark questions with 91.2% accuracy, according to a Nature paper published September 16 by Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard and James Zou.

The system uses agents to package a manuscript's supplementary materials, datasets, code and workflows into a Model Context Protocol (MCP) server, then tests it. Of 599 proposed tools generated from the 100 papers, 593 passed validation.

The authors state the thesis flatly: "Paper2Agent transforms research output from passive artefacts into active systems that accelerate use and discovery." They frame it as treating manuscripts as "active, agent-native knowledge rather than static text."

Individual cases are sharper than the aggregate. An AlphaGenome agent, generated in about 45 minutes at $14 in compute, hit 98.7% accuracy on tutorial queries and 100% on novel ones. A Scanpy agent reproduced human researcher workflows across seven single-cell datasets, with runtime improvements of 1.9 to 3.8× against baselines.

The paper also stages a small discovery experiment: multiple paper agents collaborated autonomously and pointed to GPR137 as a probable causal gene for psoriasis at the rs887314 locus, alongside prior candidates including SORT1, CELSR2 and PSRC1. Five researchers we track posted the link when the paper went up.