Skip to content
Documentation /Start

What is NosoGraph?

NosoGraph is open-source biomedical research software for connecting disease knowledge, typed claims, evidence, and source provenance across local data and upstream resources. This page is conceptual: it explains the objects and boundaries before you install anything.

Research use only. NosoGraph is not a diagnostic system, clinical decision-support system, or replacement for source literature. An association is not automatically causation, and a claim is not proof.

The core objects

A normal inspection path is:

condition or disease
        -> typed claim
        -> evidence direction
        -> study or source
        -> provenance and snapshot context
  • A condition is a disease or biomedical entity identified in the repository or an imported ontology snapshot.
  • A typed claim records a subject, predicate, and object, such as a condition HAS_PHENOTYPE relationship.
  • Evidence describes the records attached to a claim and preserves directional semantics: SUPPORTS, CONTRADICTS, INCONCLUSIVE, or UNASSERTED.
  • Provenance records the source, import context, snapshot information, and stable identifiers where available. It supports traceability; it does not establish that a scientific conclusion is correct.

What you can do locally

  • Use the nosograph CLI to inspect and validate disease modules.
  • Run the FastAPI dashboard and open the read-only Evidence Explorer.
  • Compare two through five conditions with explicit KNOWN_ABSENT and NOT_RECORDED states.
  • Use the API and structured biomedical store for programmatic research workflows.

NosoGraph complements upstream resources such as MONDO, HPO/HPOA, PubMed, ClinicalTrials.gov, Open Targets, and GWAS Catalog. Source integration maturity is tracked separately from scientific quality; see Data sources.

Current scope

NosoGraph is Public Alpha. The CLI is the most stable task-oriented surface; the API and dashboard are Beta (local/self-host only — not GitHub Pages), Evidence Explorer is Public Alpha, Compare and Evidence Workspace are Beta, and Open Targets synchronization plus optional LLM enrichment are Experimental. A public hosted demo is not deployed. DEMO_MODE exists as a default-off opt-in. FHIR, OMOP, and Phenopackets are not implemented. The package is not on PyPI.

The registry has 10,404 discoverable modules; harvest JSON and on-disk directories are different counts. Registry breadth is not curation depth. Public L2=88 and L3=2 are an n=500 sample. See Current status and coverage.

Continue