Shipped the WeDaita Biomedical Knowledge Graph (WBKG)

We just shipped the WeDaita Biomedical Knowledge Graph (WBKG), powered by Neo4j and Adaptive Graph RAG — moving beyond retrieval toward biologically grounded reasoning.

WBKG combines heterogeneous biomedical knowledge, ontology grounding, and graph-based reasoning to support context-aware AI agents for drug discovery.

Core scale:
- 190,531 nodes across 10 entity types
- 21,813,816 edges across 35 relation types
- 65 biomedical data sources grounded in 18 ontologies

Built on Harvard Medical School’s OptimusKG v2.0 from the Zitnik Lab. Thanks to Prof. Marinka Zitnik and the team for advancing open biomedical knowledge graphs.

Scale means little without precision. We address key reasoning challenges with Adaptive Graph RAG, drawing inspiration from early work by Tomaz Bratanic and PhD Fanghua (Joshua) Yu.

📍 Local — entity-level precision
🌐 Global — pathway and landscape insights
🔄 Hybrid — full-context reasoning

This is also deeply personal for me. At Pfizer, I was among the earliest adopters building AI systems that combined knowledge graphs with agent-based workflows for biomedical decision-making — shaping how I think about real-world gaps in drug discovery systems.

This is just the beginning. We’re continuing to expand WBKG with additional biomedical knowledge layers and agentic workflows.