AI Tool Turns Research Papers Into Smart Assistants That Can Answer Your Questions

Scientists have built a new artificial intelligence tool that can turn any research paper into a smart AI assistant. The tool is called Paper2Agent, and it was described in a study published in the journal Nature. It was developed by computer scientist James Zou and his team at Stanford University in California. The assistant can answer detailed questions about a paper, run tests using the paper's methods, and even team up with other AI assistants from different fields of science.

Paper2Agent works by reading everything in a research paper — the main text, the data, the computer code, and more. It stores all of this information on a digital hub called an MCP server. Then a team of AI agents automatically builds special tools that can apply the paper's methods to brand-new data. After that, scientists can connect to the server using a large language model, which is a type of AI that understands plain conversational language. This lets them talk to the assistant just like they would chat with a real expert.

Zou's team tested the tool on a paper about AlphaGenome, an AI model that predicts how DNA sequences affect genes in our bodies. Paper2Agent built a working assistant for that paper in about 45 minutes, and it only cost $14 in computing power. The assistant did extremely well on its first test, answering genetics questions with near-perfect accuracy. It also scored higher than other top biomedical AI tools that were given the same paper and asked the same questions. One of those tools, called Biomni, pulls from dozens of databases but still scored far lower than Paper2Agent's assistant.

Zou says the assistant's success is because it deeply understands AlphaGenome's own tools and methods. The team then gave the assistant a harder challenge — figuring out why a single change in a DNA 'letter' is linked to high 'bad' cholesterol. They asked it to identify the exact gene responsible for this connection. The assistant found a gene, but it was a different gene than the one identified in the original AlphaGenome paper. Zou says that both answers are supported by the data, which shows there is still more to learn.

This difference in results actually highlights one of Paper2Agent's most useful features. Scientists can use the tool to look back at published research and question old conclusions — without having to design brand-new experiments from scratch. In the past, reading a research paper was a one-way experience; you read it, but it could not respond to you. Now, with Paper2Agent, a paper can become a 'living' source of knowledge that scientists can interact with and explore further. Researchers hope this kind of technology will change the way people share and use scientific knowledge in the future.