Open · MIT · FDA public-domain · never-launder

The drug-interaction graph the world builds together.

A free, proof-carrying drug-interaction reference. Every relationship shows the exact FDA-label sentence it came from — and when a pair isn't in the label, it abstains out loud rather than guess a pair is safe.

Contribute an interaction
Every relationship cited to its FDA label · never-launder
Loading a real cited interaction…
How you can help

Add a cited interaction

A form, not a pull request — but every relationship ships with its FDA-label citation and a knowledge test. No PR required.

Contribute →
What makes OpenDrugGraph different

Never-launder

An answer carries its cited FDA-label span, or it abstains. Zero fabricated relationships — enforced by the same gate that builds the graph.

The citation is the answer

Every relationship is its proof: the SPL SET_ID, the LOINC section, and the verbatim label sentence — auditable, not a black box.

Open & verifiable

MIT over FDA public-domain labels; export CSV / JSON-LD / FHIR / RDF. Regenerated from tests — yours to check.

Powered by NuSy

A neurosymbolic engine: an LLM proposes, a symbolic gate disposes, and only proof-carrying relationships are served.

Research preview — not medical advice. MIT · FDA public-domain · never-launder: every relationship cited to its FDA label, or it abstains.
Interaction neighborhood

Search a drug to begin

Search a drug above — its cited interaction neighbors expand here. Click a relationship to see its proof.
  • Interaction
  • Contraindication
Relationships — list view
  • Search a drug to list its cited relationships.
Click a relationship to reveal its FDA-label proof: SET_ID, LOINC section, and the verbatim source span.
Flagship — prescribe-time DDI safety, cited or abstained

Medication Safety Check

A guideline recommends a drug — but is it safe against what the patient already takes? Enter a candidate drug and the patient's active medications. A cited interaction withholds the recommendation with its FDA-label proof; an unknown drug abstains loudly; no cited interaction recommends (open-world — “no cited interaction found”, never “proven safe”).

Names resolve via the schooled graph; unresolved names are reported, never guessed.
Enter a candidate + active meds and run the check — the cited disposition (Withhold / Recommend / Abstain) renders here.
Product quality — measured over the live graph

Measures

loading

Review queue

Community contributions awaiting confirmation. Two independent confirmations mark a contribution ready to promote — and the citation is the gate: a contribution with no source is Blocked and can never enter the graph.
Loading the ratification queue…

Recent changes

Every submission and status change, in the open — like a wiki, not a git log. Pick a relationship to see its full timeline.
Site-wide feed
Loading recent changes…
Per-relationship timeline
Pick a relationship from the feed — or open one in Explore and follow “View this relationship’s history” — to see its full provenance timeline.
How it works · powered by NuSy

Not a maintained table — a live reasoner over a proof-carrying graph.

Every OpenDrugGraph answer is computed by a neurosymbolic engine: a language model proposes, a symbolic gate disposes, and the result is either Proven with a full provenance trace, Heuristic (clearly labelled), or a loud abstention. The same open-source engine — NuSy — runs guideline CDS, terminology, and any domain where a wrong answer is unacceptable.

The request path — propose → dispose → prove
1 · LLM proposes
extracts candidate interactions from FDA labels — never trusted on its own
2 · Symbolic gate disposes
must cite + ground + compile, or it is flagged · false_assert=0
3 · Reasoner proves
mechanism-DAG derivation over Arrow — the 2-hop path is the proof
4 · Proven · Heuristic · Abstain
every answer carries its status + provenance — the reasoner cannot lie

The engine, by mechanism

Each is real, open code — not a slogan.
Proof-carrying reasoning

An LLM proposes; a symbolic gate decides. Every answer is Proven / Heuristic / Abstain, computed from a proof trace — never set. nusy-reasoner · Provability

Never-launder gate

No relationship enters the graph without a complete FDA-label citation + verified codes, or it is flagged. nusy-drug-graph · emit::gate_edge

Mechanism-DAG derivation

drug→enzyme/transporter links + forward-chaining derive interactions a pair-table cannot hold. nusy-forward-chain · proof.rs

Arrow-native store

The graph is columnar Arrow / Parquet — fast to query, and exportable as CSV, JSON-LD, FHIR, and Turtle. nusy-arrow-core

Knowledge-tested (KBDD)

Every relationship ships a Given/When/Then competency question and promotes only by passing it. nusy-kbdd · competency_question

Provenance ledger

Every change — by anyone, any role — is cited, logged, and reversible into review. nusy-provenance-ledger

Every answer is a data structure you can inspect

Representative response shape — GET /api/edge/:id over the live graph.
GET /api/edge/:id { "pair": "Clarithromycin ⇌ Simvastatin", "verdict": "Proven", "kind": "Mechanism (2-hop DAG) + cited label", "proof": { "premises": [ "clarithromycin CYP3A4", "simvastatin substrate_of CYP3A4" ] }, "citation": { "set_id": "a7f3b0c9", "loinc": "34073-7", "span": "…contraindicated…rhabdomyolysis." }, "false_proofs": 0 }
Powered by NuSy — the reasoning platform

OpenDrugGraph is one instance of a general engine.

The same proof-carrying, never-launder reasoning runs guideline CDS, clinical terminology, and any regulated-domain decision support. The reasoners are MIT-licensed and open — build on them, or have us build your domain.

AI-review report · where you can help

Needs work

The AI proposes cited interactions from FDA labels; humans verify them and fill the gaps. These are the drugs whose must-not-miss interactions are still missing — highest-harm gaps first.

The prioritized queue

avoid/modify gaps first, least-complete first · click a column to re-sort
Loading…
Community · the coverage gap, in the open

Wanted List

The most-prescribed drugs that have no cited interaction yet, in prescription-volume order. Pick one and add a cited interaction — every contribution shrinks the gap toward the coverage north-star.

Loading…
Data · open, MIT, download the whole graph

Download the graph

The entire proof-carrying drug-interaction graph, free and MIT-licensed, in machine-readable formats. Every export carries only cited edges — the never-launder honesty tally (skipped_uncited) and the licence tally (skipped_license) are shown per format so you can audit the open-data claim yourself. Browsing is open; downloading requires a free sign-in.

Loading formats…

Reference catalog

Cite & verify

A citable dataset DOI (Zenodo) and downloadable OWL/SHACL shapes for self-verification are planned — until an artifact is actually published we show it as coming, never a dead link. See /api/download/formats for the live, authoritative manifest.