Ask biology.
Get sourced answers.
RamonY is a field-scale knowledge graph — a “mundaneum” — of mammalian neuroscience, built and operated at EPFL. It extracts structured facts from the neuroscience literature, databases and datasets, and records on every relation the conditions under which it was observed, including species, cell type, developmental stage, brain region, direction of effect and assay. The service is aimed at academic neuroscientists and life scientists.
- Nodes
- 138M
- Edges
- 276M
- Databases
- ~370
- Edge types
- 21
How it works
Ask in plain English
Type a biology question. An AI assistant translates it into a graph query, executes it against the live database, and interprets the results with full source citations.
Open →Explore visually
Search for any gene, cell type, brain region, or disease. Its neighborhood appears as a 3D graph — click any node to expand and discover connections across hundreds of databases.
Open →Try these questions
A small selection of questions the graph can answer right now. Each runs against the live database. Click to open it in the Ask interface.
What's inside the graph
Sources are grouped by the role they play: foundation ontologies define the entity vocabulary; curated and focused databases provide high-confidence relationships; quantitative resources add per-gene and per-cell numeric measurements; structural data adds residue-level contact maps. Counts below show live / total declared.
UBERON, Cell Ontology, Gene Ontology, ChEBI, MONDO, HPO, MP, ATC, EFO, NCBI Taxonomy, Sequence Ontology, PRO, PATO, OBI, OAE, the Allen brain atlas and several developmental-stage and modification ontologies — these define the canonical entity classes. Island-linker derived edges (ATC + Linnarsson) included.
HGNC, UniProt, RefSeq, STRING, IntAct, DisGeNET, Reactome, ChEMBL (including 2.29M bioactivity edges now source-attributed), DGIdb, OncoKB, Orphanet, HMDB, FooDB, SMPDB, CORUM, PhosphoSitePlus, MitoCarta, STITCH — the core identifier registries and curated relationship sets.
Allen ABC, BRENDA, BioGRID, CATlas human brain (7.2M cCRE–CellType edges), ComplexPortal, CRISPRbrain ORCS, ENCODE, HPA, InterPro, Lister methylome (1.8M Gene→CellType), MGI PhenoGenoMP (300K), miRTarBase, Open Targets full evidence (3M), PhaSepDB, RNAcentral, SapBERT-similar (66K), SignOR, SynaptomeDB (525K), SynGO, TarBase, TCDB, ReMap, TRRUST, NeuroSynth, NeuroMorpho, neurodev qualifiers — and over a hundred more focused resources.
GTEx expression and eQTLs, HCA cell-type avg expression (47K edges, 424 CL: types), CellxGene Census full (843K edges), mtDNA tissue expression, HPA single-cell / brain / tissue, BrainSpan, AlphaMissense, REVEL, SpliceAI, ReMap2022 TF binding, ClinVar, DepMap dependencies, CellMarker, OpenTargets, FinnGen, UK Biobank, rfam_gene_annotations + super_enhancers fixes (+2,764 edges), plus full GENCODE annotations and per-feature genomic statistics.
Braun, Velmeshev, Bhaduri, Nowakowski, Polioudakis, Eze, Micali and Sundström — cell-type expression and chromatin atlases of the human developing brain integrated as cell-type / region / stage edges.
RNA residue graphs are loaded; PDB residue graphs are staged. These power structure-aware queries about specific binding pockets and interfaces.
Every edge in the graph carries a hierarchy_source qualifier identifying the database it came from. Click a source name in any result to open the corresponding database.
One shared schema for every database
Each database, no matter how it represents its data internally, is rewritten into a small vocabulary of entity classes (gene, protein, cell type, anatomical region, disease, drug, …) and the relationship types listed below. The original biological detail is never lost: it travels with the edge as a structured set of qualifiers.
Every edge keeps its biological nuance
Relationship-type names alone (e.g. “expressed_in”) lose too much detail. Every edge in the graph carries a structured set of qualifiers that preserves the experimental and biological context. The cards below illustrate the qualifier types the graph captures — the kind of structured context attached to each edge. Sign in to pull live examples straight from the graph.
Live edge examples require sign-in
The graph API is protected. Sign in to view live qualifier examples, run Cypher, and explore graph neighborhoods.
Sign inQualifier categories at a glance
The structured layer behind those examples — these are the canonical qualifier types, each appearing on the relevant edge classes.
Every edge cites the database it came from. Click any source on a result to open the upstream DB.
Whether the observation is human, mouse, or cross-species — never assumed.
Binding, inhibition, agonism, activation… how a causal influence is actually exerted.
Whether the subject increases or decreases the object — never lost in “associated with”.
Numeric expression value carried in TPM, nTPM, FPKM, z-score or HPA ordinal levels — preserved alongside the edge.
STRING's 0–1000 confidence, BindingDB Kd, IntAct score — kept in native scale.
Residue, position and modification type for every PTM (PhosphoSitePlus, dbPTM, PSI-MOD).
ClinVar clinical_significance + AlphaMissense / REVEL / SpliceAI numeric scores side by side.
The way one anatomical region contains another — sub-region, layer, lobe, leaflet, compartment.
Physical vs genetic interaction — disambiguated for every PPI/genetic-screen edge.
GWAS p-value, DisGeNET score, eQTL FDR — preserved verbatim from the source.
The specific verb within an edge type — “binds”, “activates”, “phosphorylates”, “projects to”.