Stanford's Paper2Agent turns AlphaGenome paper into an agent

Paper2Agent, a tool from Stanford computer scientist James Zou and his lab, built an autonomous AI agent from the AlphaGenome paper in about 45 minutes for $14 in compute, Nature reports. The agent then answered genetics questions with near-perfect accuracy and outscored Biomni, a biomedical AI tool built by academic researchers that draws from dozens of databases.

The pipeline is straightforward. Paper2Agent ingests a paper's main text, code, datasets and supplementary materials, and, per Nature, "the information is deposited onto a digital platform called an MCP server." From there, AI agents build tools that apply the paper's methods to new data, and a scientist can query them through their preferred large language model.

The AlphaGenome test produced one striking result. Asked why a single change to a DNA "letter" is associated with "bad" cholesterol, the agent named a different gene than the one pinpointed in the original paper. In Nature's telling, it "identified a gene ... a different causal gene to the one pinpointed in the original AlphaGenome paper."

Zou frames the ambition broadly. The approach, he says, "can help us to reimagine what knowledge looks like in the future."

Nature's write-up tests the tool on one paper. Three of the researchers we track shared it the same day.