Reproduce “DeepEvidence: Empowering Biomedical Discovery with Deep Knowledge Graph Research” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zifeng Wang, Zheng Chen, Ziwei Yang, Xuan Wang, Qiao Jin, Yifan Peng, Zhiyong Lu, Jimeng Sun

DeepEvidence, an AI-agent framework from Keiji AI and NIH, conducts "Deep Research" across heterogeneous biomedical knowledge graphs, addressing the limitations of text-only AI systems. The framework achieved substantial performance improvements on established biomedical benchmarks and new tasks across the discovery lifecycle, such as 40.0% on HLE-Medicine, 80.0% on LabBench-LitQA2, and 68% for Drug Discovery Target Identification.

Key insights

  • A structured, agentic approach combining broad and deep exploration strategies is effective for navigating and bridging heterogeneous biomedical KGs beyond traditional text-based reasoning.
  • Explicitly maintaining an internal, continually updated evidence graph is crucial for performing deep, targeted, and reliable knowledge graph investigation and comprehensive evidence synthesis.
  • Integrating unified access to a diverse array of specialized biomedical KGs, coupled with an execution sandbox, enables AI agents to execute complex, multi-step queries and data analysis, addressing the limitations of general-purpose LLM tool use.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.