Turning a paper into a conversational agent: reading the Paper2Agent report
Synopsis
According to a Nature news report, Paper2Agent reads a paper's main text, code and data sets, deposits them on an MCP server, and has a team of AI agents autonomously build tools that apply the paper's methods, producing a paper-specific agent that can be questioned in plain language as a 'virtual corresponding author'; the report says it created an agent for the AlphaGenome paper in about 45 minutes at US$14 of computing cost, that the agent answered genetics questions with near-perfect accuracy and outscored other top biomedical AI agents including Biomni, and that it was used to re-examine which causal gene explains a single-letter DNA change linked to 'bad' cholesterol, arriving at a different gene from the one pinpointed in the original paper.
Interpretation
The report says Paper2Agent converts a static paper into a paper-specific agent that answers questions immediately, acting as a 'virtual corresponding author'. Compared with manually digesting a paper or simply pointing a general model at it, the tool automatically places the paper's main text, code and data sets on an MCP server and then has a team of agents build tools that can apply the paper's methods. Based on the Nature news description of the workflow and comments from co-author James Zou; the report does not provide methodological detail or statistics from the peer-reviewed paper.
The report says the tool autonomously created an agent for the AlphaGenome paper in about 45 minutes at US$14 of computing cost, that it answered genetics questions with near-perfect accuracy, and that it scored higher than other top biomedical AI agents given the same paper, including Biomni. This offers a comparable instance: the same paper and the same questions, with a paper-specific agent versus general biomedical agents. Based on test results and author statements relayed by the report; the report does not list the number of questions, the scoring rubric or margins of error.
The report says the team asked the agent to investigate why a single DNA letter change is associated with 'bad' cholesterol, and the agent identified a causal gene different from the one pinpointed in the original AlphaGenome paper. This illustrates a new use of paper-as-agent: re-evaluating published conclusions without designing entirely new experiments. Based on the report's narrative; Zou says AlphaGenome's genetic variant data support both hypotheses, and the report gives no further validating experiment.
Perspective
The report speaks to researchers who want to grasp advances in unfamiliar fields quickly or to reuse a paper's methods directly; it applies where a paper has accessible main text, code and data sets and can be connected to an MCP server and a chosen large language model. The example given centres on AlphaGenome, an AI model that predicts properties of DNA sequences, and its genetics question-answering, so the reported conclusions mainly apply to that setting; the report also mentions that agents can collaborate autonomously with agents for papers from other disciplines, but does not detail specific disciplinary cases.
The report is news relay and does not include the cited paper in full, so the number of test questions, the scoring method, the comparison conditions and statistical uncertainty remain unclear; on the finding that the agent's causal gene differs from the original paper's, the report only relays the author's statement that data support both hypotheses, which awaits independent verification; the report also mentions autonomous cross-disciplinary collaboration between agents without describing how that collaboration performed or where it applies.