Knowledge graph: Overview

The Cancer Complexity Knowledge Portal lists datasets, publications, tools, grants and educational resources as five separate tables. Answering a question that spans more than one of them — which tools were used on datasets about a given disease, what a consortium's funding produced — normally means joining those tables by hand. This pipeline builds the same information into one graph instead, so a single query can answer it.

This page is the plain-language overview. The other pages under Knowledge Graph in the navigation cover the technical depth, and the kg-pipeline README has the full command reference.

1. What the graph is for

The graph resolves each portal value to a real scientific term where one exists, and links related records together — a dataset to its funding grant, a publication to the dataset it describes. It never guesses: a value with no match stays as plain text.

Most links come straight from matching values. Its Program and disease/tissue crosswalks into the public graph wait for a person to approve them; the private file-level layer links to sagebrain-model automatically instead, on an exact label match only.

What it deliberately doesn't do:

It doesn't	Because
Force a match for every value	An unmatched value stays as plain text, not a guess
Apply a public-graph crosswalk on its own	Program and disease/tissue rows wait for approval (see Federation)
Assert the access-control graph's own facts	That graph owns those facts; this one only shares an identifier with it
Extract the portal's own Person table	Not pulled in yet; investigators and contributors become provisional stubs instead
Mix in the private, file-level layer	It's built and published separately

2. Where it sits

Part	Role
Synapse	The source: portal tables, dataset annotations, and separately, file-level annotations
This repo's model	Supplies the vocabulary the graph resolves values against
The pipeline	Builds, checks and publishes the graph — see Build and validation
The published graph	Available from Synapse, and from Sage Bionetworks' shared graph store
Federation targets	Links to the Sage Common Data Model and to ontology crosswalks — see Federation
The private layer	A separate, access-controlled graph of file-level facts
The access-control graph	A sibling system that shares only an identifier with this one

3. The layers

The graph is built in layers: a schema, one record per portal row, enrichment that adds real terms and a second type, and federation links to sibling systems. A separate, private layer covers file-level facts. Details are in Model and identifiers and Build and validation.

4. Last verified build (2026-09-29)

Rows extracted	1141 datasets, 4773 publications, 349 tools (331 nodes), 160 grants, 10 resources, 966 catalog rows
Graph size	304,989 triples in the public portal graph; 337,917 once federation is added
Deposited to the shared graph store	360,370 distinct triples

Part	Status
Portal graph, validation and query checks	Operational; all checks pass on the live build
Organization links to the Sage Common Data Model	Operational; minted automatically from each institution's registry id
Program links	Operational; all 11 consortium rows are approved
Disease/tissue crosswalks, public graph	Built and tested; no rows approved yet, so no links
Disease/tissue crosswalks, private layer	Operational; applies automatically on an exact label match
Private file-level layer	Implemented; published only to an access-restricted location
Deposit to the shared graph store	Operational; latest deposit 2026-09-29

Where to look next

For	See
Schemas, identifiers, and a worked example	Model and identifiers
The build pipeline and validation	Build and validation
Publishing and the deposit guards	Publishing and deposit
A query cookbook	Querying the graph
Federation with sibling systems	Federation and sagebrain-model
The full command reference	the kg-pipeline README
Design rationale and scope limits	the architecture decisions plan
The data model itself	Home and the Data Models section