DeepEvidence AI Agent

All agentsPaper

Hierarchical deep-research agent orchestrating specialist sub-agents across biomedical sources.

This page explains what DeepEvidence is, who should use it, and why it may be the right agent for your workflow.

DeepEvidence is a biomedical research agent built to explore many specialized knowledge sources together, especially biomedical knowledge graphs that connect genes, diseases, drugs, pathways, and clinical evidence. It combines broad search across multiple resources with deeper multi-step reasoning to gather, organize, and synthesize evidence into a structured view. This helps researchers investigate complex questions more systematically across the full discovery pipeline, from early drug discovery to clinical research and evidence-based medicine.

Examples

QUESTION

🕶️ Do blue-light filtering glasses improve sleep in healthy adults? Use only human studies and prioritize systematic reviews/meta-analyses over single trials. Return a table with: study type, sample size, effect direction, effect size (if reported), and PMID/DOI + direct link for every claim.

EXPECTED ANSWER

Blue light blocking glasses did not improve objective measures of sleep time or quality in healthy adults.

QUESTION

🌿 Why does cilantro taste soapy to some people? Summarize evidence with sources for: 1) the main odor molecules implicated, 2) the top associated human gene, 3) one GWAS citation (with PMID), 4) a short caveat on association vs causation.

EXPECTED ANSWER

Should identify OR6A2 as the commonly implicated gene, mention cilantro-associated aldehydes (for example, (E)-2-decenal), cite GWAS PMID 22828934, and include a caveat that GWAS associations do not by themselves prove causation.

QUESTION

🧠 For Alzheimer's disease, which protein targets have the strongest non-clinical support for drug development based on (1) pathway involvement, (2) genetics evidence, and (3) preclinical model results? Show evidence by source database and publication.

EXPECTED ANSWER

Should stay in discovery/preclinical territory (not treatment advice). The response should rank candidate targets and cite verifiable evidence by source database and publication for pathway links, human genetics support, and preclinical model findings.

Details

  • License: MIT
  • Citation: Wang et al. (2025). DeepEvidence: Empowering Biomedical Discovery with Deep Knowledge Graph Research. arXiv preprint.
  • Models: gpt-5-mini