Crosswalks

Shared identifiers and mappings that link the OKN knowledge graphs

A crosswalk is a verified way to join two knowledge graphs on a shared identifier — so a question that no single graph can answer becomes answerable by combining them.

Explore the crosswalks

What is a crosswalk?

A crosswalk is a verified way to join two (or three) knowledge graphs on a shared identifier — for example, linking a disease in one graph to the genes another graph associates with it through a common MONDO or DOID code.

The OKN graphs are built by different teams, in different domains, on different ontologies. Each one is useful on its own, but the value of the federation is in the connections between them: a crosswalk is an integration opportunity, a place where a question that no single graph can answer becomes answerable by combining two.

“Which genes are associated with autism spectrum disorder, and what is known about them?”

A rare-disease graph knows the gene–disease associations. A biomedical graph knows the pathways, drugs, and protein interactions for those genes. Neither can answer the whole question — the crosswalk on the shared disease identifier joins them into one result.

“What is the maximum PFAS measurement in each county?”

A contamination-monitoring graph holds the measurements at sampling sites. A geospatial graph holds county boundaries and their identifiers. The crosswalk on the shared place identifier aggregates the readings up to counties.

How they work

Every crosswalk names a shared key — the identifier that both graphs happen to use for the same real-world thing — along with the predicates that carry it on each side and the number of rows the join actually returns.

• MONDO / DOID disease identifiers
• Entrez / HGNC / Ensembl gene identifiers
• CHEBI / PubChem chemical identifiers
• NCBITaxon organism identifiers
• FIPS / GEOID geographic area identifiers

Not every pair lines up directly. Where two graphs use different schemes for the same entity, the join is routed through a bridge graph — ubergraph maps DOID to MONDO, for instance, and wikidata maps HGNC to Entrez. The network map shows these as dashed and dotted edges so you can see at a glance which connections are direct and which are bridged.

Crosswalks are verified, not inferred. Each one has been run as live federated SPARQL against the endpoints to confirm it returns rows, and the findings were checked against an authoritative source — biomedical claims against the published literature, geospatial and industrial joins against the shared standard they key on.

How they are used

Crosswalks are the routing layer for questions that span the federation. Rather than reading 35 schemas to work out where two graphs might meet, you look up the verified recipe and write the query against it.

They are also what lets an AI assistant query OKN reliably. The MCP server exposes the crosswalk table as tools an assistant calls before writing any SPARQL: one lists every verified integration point across the federation, another returns the full join recipe for a specific pair of graphs, and a third answers the reverse question — which graphs can supply pathway, trait, or disease context for an entity you already have an identifier for. Because a graph pair that has been checked and does not join is recorded as such, the assistant knows which paths are dead ends instead of guessing.

What is in the catalog

Every crosswalk is worked twice — two research questions each — and every question links to a standalone, replayable transcript containing the prompt, the answer, and every verbatim SPARQL query with its results.

The catalog covers 15 domains: