Paper2Agent converts 74 biology papers into working AI agents

A Stanford-led team ran 100 computational biology papers through an automated pipeline and turned 74 of them into working AI agents without manual cleanup, according to a paper published in Nature on September 16.

The framework, called Paper2Agent, converts a paper's methods into a Model Context Protocol (MCP) server so its tools, resources, and workflows can be queried in natural language. On the AlphaGenome paper alone, authors Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan K. Pritchard, and James Zou report that the pipeline generated 22 validated tools in 45 minutes at a cost of $14, hitting 98.7% accuracy on tutorial-derived queries and 100% on novel ones. Across the 100-paper set, 593 tools passed automated validation and the resulting agents answered benchmark questions at 91.2% accuracy.

The paper describes the shift as one "from a document or codebase encoding knowledge to a knowledgeable entity capable of execution and dialogue."

In one worked example, three separate paper-agents (AlphaGenome, MPRA-coupled scCRISPRi, and Perturb-seq) autonomously interacted to identify GPR137 as the probable causal gene at the rs887314 locus in psoriasis.

Roughly 26% of papers failed agentification, the authors report, typically because code was missing, documentation was incomplete, or environments could not be resolved. The paper proposes journals begin adding an "Agent availability" section alongside the usual data and code availability statements. Five of the researchers on our Who's Who list posted the link the same week.