BrainKB
AgentPubEvidence
v2.0 · NIH Reach Tools

The open
neuroscience
knowledge graph.

Browse agents, evidence, projects, and publications across a unified, LLM-ready graph. Query with natural language, SPARQL, or the visual explorer.

73,060
RDF triples
5,811
Entities
12,847
Evidence assertions
3,241
Publications linked
Powering research at
Allen InstituteNIH BRAINStanford NeuroEBIJaneliaMIT McGovernUCSFMax Planck
Architecture

Four layers. One trustworthy graph.

01 / 04

Agents & extraction

Multi-agent LLM pipeline extracts entities, relations, and provenance from papers, datasets, and structured sources.

02 / 04

Evidence & provenance

Every assertion linked to source text, version, and curator. Confidence scoring and conflict resolution built-in.

03 / 04

Knowledge graph store

RDF triples on an open SPARQL endpoint. Neuro-specific ontologies aligned to NIMH Data Archive, UBERON, and NCBI.

04 / 04

NIH Reach toolkit

Cross-referenced views for funding officers, PIs, and the curious public. Export-ready for reporting.

Cross-linked by design

Follow the thread from agent to evidence to publication.

Every entity shows its incoming and outgoing edges. Click an assertion to see the extracting agent, the source paragraph, and the paper it came from — with provenance baked into every node.

Agentbkb:agent/PV-interneuron-CA1 2.4k views

PV+ interneuron (CA1)

Fast-spiking parvalbumin-positive GABAergic interneuron · Hippocampal region CA1
47
Evidence
23
Publications
8
Datasets
Recent assertions
expressesMarker→ParvalbuminKepecs 2014
locatedIn→stratum pyramidalePelkey 2017
firingRate→40–200 HzHu 2014
Explore BrainKB

Four ways in.

How it works

From paper to triple, with provenance at every step.

01
Ingest
Papers, datasets, and structured sources enter the pipeline.
02
Extract
Agents propose assertions with source spans and confidence.
03
Review
Curators accept, edit, or reject. Conflicts are surfaced.
04
Publish
Accepted triples hit the graph with full provenance.