VaidhyaMegha/regulatory-affairs-kg

Medical-Device Regulatory Affairs Knowledge Graph

28,520 nodes. 25,332 edges. The device-to-law backbone of US medical-device regulation: every 510(k) clearance for cardiovascular devices, the product code each was cleared under, and the Code of Federal Regulations section that governs it.

Built with Samyama Graph. Loader and ETL: samyama-ai/regulatory-affairs-kg.

The question it answers

When a rule or a recognised standard changes, a manufacturer has to work out fast which of their devices are affected. That is a traversal — device → product code → regulation — and it is the path this graph makes queryable:

// Everything cleared under a given CFR section
MATCH (s:Submission)-[:CLASSIFIED_AS]->(p:ProductCode)-[:GOVERNED_BY]->(r:Regulation)
WHERE r.cfr_section = '870.1130'
RETURN s.applicant, s.device_name, s.date_received
ORDER BY s.date_received DESC

// Which regulations carry the most cleared devices — where a change hurts most
MATCH (s:Submission)-[:CLASSIFIED_AS]->(p:ProductCode)-[:GOVERNED_BY]->(r:Regulation)
RETURN r.cfr_section, count(DISTINCT s) AS clearances, count(DISTINCT p) AS product_codes
ORDER BY clearances DESC LIMIT 10

Licence — all sources are US Government public domain

Source	Publisher	Terms	Feeds
openFDA 510(k)	FDA	Public domain (US Gov)	Submission
openFDA classification	FDA	Public domain (US Gov)	ProductCode, Regulation

Nothing here restricts redistribution. license: other is used because no SPDX tag says "US Government work outside copyright"; openFDA's own terms govern.

No standards text is included, by design. The upstream schema is explicit — "Standard — identifier and clause references ONLY. ISO/IEC texts are" not stored. ISO and IEC standards are copyrighted and sold; naming ISO 10993-1 as a designation is not reproducing it. The repository also designs a Clause layer holding paragraph-level obligation text from eCFR and EUR-Lex, but that layer is not loaded and is not in this dataset.

FDA does not endorse this dataset or anything derived from it.

⚠️ Scope — this is a bounded slice, and the boundary is specific

Layer	Coverage
Regulation — 2,286	All CFR parts — 866, 862, 870, 868, 882, 886 and the rest. The complete classification backbone.
ProductCode — 7,088	All device classifications, every part
Submission — 19,146	21 CFR Part 870 — Cardiovascular Devices only

Every one of the 19,146 clearances resolves to part 870. There are 175,686 510(k) clearances at the source; this dataset holds the cardiovascular subset. Do not read counts here as national totals for all device types.

Predicate chains — the "which device was this cleared as equivalent to" links that make 510(k) interesting — are designed in the schema but not loaded: the source data is not in the FDA's API, it is in the 510(k) Summary PDFs.

Files

File	Rows	Columns
nodes/submission.csv	19,146	id, advisory_committee, applicant, clearance_type, country_code, date_received, decision, …
nodes/productcode.csv	7,088	id, definition, device_class, device_name, extraction_method, implant_flag, life_sustaining_flag, …
nodes/regulation.csv	2,286	id, cfr_section, extraction_method, part, retrieved_at, section, source
edges/classified_as.csv	19,146	Submission → ProductCode
edges/governed_by.csv	6,186	ProductCode → Regulation

Usage

from datasets import load_dataset
clearances = load_dataset("VaidhyaMegha/regulatory-affairs-kg", "submission", revision="v1.0")

Verification

• Totals reconcile against the snapshot header: 28,520 nodes, 25,332 edges.
• 0 dangling edges.
• 902 orphan ProductCode nodes, and this is correct rather than a defect: the loader reports "902 product codes carry no regulation number (unclassified or exempt) — no GOVERNED_BY edge." An exempt device genuinely has no governing CFR section, and inventing one would be worse than leaving it unlinked.
• Reproduces real FDA structure: the highest-volume product codes come out as non-invasive blood-pressure systems (1,221), percutaneous catheters (895), guide wires (766) and oximeters (726) — which is what the 510(k) register actually looks like for cardiovascular devices. Clearance dates run 1976 → 2026, and 1976 is when the Medical Device Amendments took effect.
• Round-trip verified after upload.

Limitations

• Built from a live API, so counts drift. This build came out at 28,520 nodes / 25,332 edges against the repository's recorded 28,496 / 25,310 — the FDA gained records between their measurement (2026-08-12) and this one (2026-08-29). Nothing is wrong; a live source is simply not reproducible by date alone. Cite the revision.
• decision is empty on every Submission. The field exists in the schema but the API slice used does not populate it. Do not filter on it.
• Cardiovascular clearances only — see Scope.
• No predicate chains, no adverse events, no recalls, no UDI, no registrations. The upstream repo measures 8 openFDA endpoints totalling 31.1M source records; this loads two of them.
• Two device descriptions were altered during load to be representable — both contained mixed quote characters. The loader reported this rather than silently dropping them.

Not regulatory advice. For any actual submission, use the FDA's own databases as the authoritative record.

Citation

Medical-Device Regulatory Affairs Knowledge Graph, v1.0 (28,520 nodes, 25,332 edges).
Built with Samyama Graph.
https://huggingface.co/datasets/VaidhyaMegha/regulatory-affairs-kg
ETL: https://github.com/samyama-ai/regulatory-affairs-kg
Source: openFDA device 510(k) and classification endpoints, US Food and Drug
        Administration — public domain. https://open.fda.gov/
Retrieved: 2026-08-29. Scope: 21 CFR Part 870 (cardiovascular devices).