Reimagining research papers as interactive and reliable AI agents
9 hours ago
- Paper2Agent is an automated framework that converts research papers into interactive AI agents, enabling natural-language access to methods, data, code, and workflows.
- It uses a multi-agent pipeline to analyze a paper and codebase, extract core methods as MCP tools, and generate validated, reproducible model context protocol (MCP) servers.
- Each paper agent includes MCP tools, static resources, and reusable prompts, making complex scientific analyses accessible without programming expertise.
- Case studies with AlphaGenome, Scanpy, and TISSUE show the agents reproduce original results and handle novel queries, outperforming direct code-repository baselines in accuracy, speed, and cost.
- In large-scale evaluation, Paper2Agent successfully agentified 74 out of 100 computational biology papers and generalized to non-biology fields, with automated validation ensuring reliability.
- Paper agents can collaborate with each other, as demonstrated by integrating AlphaGenome, scCRISPRi, and Perturb-seq agents to prioritize GPR137 as a causal gene for psoriasis.
- The framework addresses reproducibility and code hallucination by locking validated tools, embedding source-code references, and including iterative test-and-refine loops.
- Paper2Agent proposes a new paradigm for scientific communication where papers become executable, interactive, agent-native research objects rather than static documents.