Every number on this page is read from the database as you load it — nothing here is a marketing figure typed in by hand. The catalog grows on its own schedule, gets fact-checked against primary sources, and is free for anyone to reuse.
Everything above is live, which is what makes it useless for a citation. A Gene Knowledge Graph Snapshot is the whole dataset on one date, frozen, hashed, and kept at a permanent address.
95,092 variants · 375 conditions · 55 drugs · 4 smell & taste · 33 topics · joined by 159,031 links · 7471 KB gzipped
| Five axes | variants · conditions · drugs · smell & taste · topics |
| The joins | variant↔condition, condition↔drug, sense↔variant, topic↔variant — the part a scrape of the pages does not give you |
| Per variant | genotype-by-genotype reading, gene, trait, sources |
| Per condition | MONDO / OMIM / ICD-10 codes, cited sources, last fact-check date |
| Format | one gzipped UTF-8 JSON file · one json.load reads all of it |
| Field names | the same ones the public API uses |
| Verification | published SHA-256 to check your copy against |
| Use | no expiry · no download limit · one version is normally enough |
This catalogue is under continuous curation — 7,594 variants have been added since this version was sealed. The working catalogue is always ahead of any sealed version, and a sealed version is the only form of it that can be quoted, audited or reproduced. If you need queries against today's data, that is API Pro rather than a file.
New findings arrive from the public GWAS Catalog throughout the day, and only what clears a fixed quality gate is published — nothing is added by hand and nothing skips the gate. What doesn't clear it yet is held, not discarded, and rechecked nightly as the catalog's own records fill in — see the variant count above. The curve is the running total; the bars underneath are what cleared it in each period.
The last 30 days, 22 September to 21 October 2026, counted on your clock (UTC). Not a selection of the flattering ones — this is what there is.
of everything that reached this site in the last 30 days
Of 18,785 visits recorded as not bots in the last 30 days, 11,008 confirmed people (14,453 views) — scrapers and the operator removed. How
someone opened the page
collecting for training and indexes, with nobody waiting
a person asked an assistant a question and it came here for the answer
2026-09-22 – 2026-10-21 · updated 2026-10-21 09:36 UTC
"Reads" — an AI company's crawler collecting pages to train a model or build the index its assistant searches. "Answers" — an assistant fetching a page at that moment because a person asked it something.
| Company | Headquarters | Reads | Answers |
|---|---|---|---|
| Amazon | US | 198,782 | 0 |
| Claude | US | 129,580 | 311 |
| Meta AI | US | 77,464 | 0 |
| ByteDance (Doubao) | CN | 36,523 | 0 |
| Perplexity | US | 33,704 | 0 |
| ChatGPT | US | 1,812 | 430 |
The full AI Visibility report — day by day, countries, languages and how it is counted →
Three things run whether or not anyone is watching, and the results — including the unflattering ones — are shown below.
New associations are imported from the GWAS Catalog every twenty minutes. Anything whose underlying allele data isn't clean is refused rather than guessed at.
Every cited PMID is checked against Europe PMC for matching author and title, every source URL for reachability, and every MONDO/ICD-10 code against its registry. Mismatches are flagged, not published.
A nightly watch flags pages whose linked variants have picked up newer research, and pages that have gone too long without a fact-check.
near RPL5P16 · rs4703918 · added Oct 21
See detailed info → StandardITGA9-AS1 · rs4678585 · added Oct 21
See detailed info → StandardSPTBN1 · rs4671217 · added Oct 21
See detailed info → Standardnear LINC00616 · rs4640713 · added Oct 21
See detailed info → StandardBNC2 · rs4626704 · added Oct 21
See detailed info → Sensitivenear CASC16 · rs4608329 · added Oct 21
See detailed info →The counts above describe five separate lists. What makes them worth having together is what joins them: 170,288 edges across 130,646 nodes, each labelled with the evidence behind it — a link a person curated from a cited source, or a co-mention computed from the literature, never the two silently added together.
One page states how an edge is counted and why, and shows what the graph leans on hardest. Every connection in it is readable here on the pages; the assembled machine-readable list is the one paid thing.
Gene Knowledge Graph →The structured data is free to take — no key, no signup, no negotiation. Most comparable sources are non-commercial, academic-license-only, or require an approved account; ours is CC BY 4.0, which means you can build on it, publish with it, and ship products on it as long as you credit us. The written sections and the links between entities are published on the pages rather than served by the API — that part is the work, and it is free to read.
curl https://www.mygenelog.com/api/v1/variants
All 102,686, 100 to a page with a next link — per-genotype interpretations and the source cited for each.
curl https://www.mygenelog.com/api/v1/conditions
Linked variants, MONDO/OMIM/ICD-10 cross-references, pharmacogenomic notes, every cited source, the fact-check date — and the open questions we could not resolve.
If you use this in a paper, a thesis, or a product, cite it like any other data source:
MyGeneLog™ (2026). MyGeneLog™ variant and condition dataset [Data set]. narubox, Austin, TX. Retrieved from https://www.mygenelog.com/data
@misc{mygenelog,
title = {MyGeneLog variant and condition dataset},
author = ,
year = {2026},
note = {CC BY 4.0},
url = {https://www.mygenelog.com/data}
}
Citing this needs a version that will still say the same thing in five years — see below. Need a field we do not expose yet, or a different cut? Ask us — we would rather build it than have you scrape it.
Every author of every cited paper is named on the citation list. If yours is missing, or a variant on this site should trace to one you wrote, send us the identifier — corrections are more welcome than new papers.
Submit a paper →The edges, not the nodes. Variant catalogues, disease ontologies, pharmacogenomic guidelines and chemosensory research each already exist as excellent open resources, and each is separate from the others. Here every variant carries the conditions it links to, every condition carries the gene-drug pairs that apply, the drug pages read the same pharmacogenomics from the other direction, smell and taste sit alongside them because they are how food gets chosen in the first place, and a topic index runs the whole thing backwards — from a word people search for to the positions that appear in the same papers. Anyone can rebuild our node lists from public sources; the joins are what took the work — which is why you will find them on the pages, free to read and to link to, rather than in the API.
The structured data, yes — variants, conditions, drug pairs, the chemosensory set, every source and review date, all CC BY 4.0 with attribution to MyGeneLog™. That is deliberate: most comparable sources are non-commercial-only, academic-license-only, or require an approved account, which blocks exactly the people who would build something useful. Two things sit outside it and the API does not serve them: the written overviews and clinical sections, and the links between entities. Those are on the pages, free to read and link to. If you need the links themselves machine-readable, tell us what you are building.
No. Both /api/v1/variants and /api/v1/conditions are open, unauthenticated JSON endpoints. A light rate limit applies (10 requests per minute per IP) because this runs on a small server, not a CDN. If you need more, ask and we will work something out rather than have you scrape.
Variants are collected from the public GWAS Catalog every fifteen minutes and published only if they pass a fixed quality gate. Condition pages are written and fact-checked individually, then re-checked on a schedule; each carries the date of its last verification. Citations and identifier codes are re-validated weekly against Europe PMC, MONDO and the ICD-10 registry.
A GWAS Catalog association our collector has read but not yet turned into a published variant page — an unclear risk allele, no gene mapping the catalogue or Ensembl’s own coordinates could confirm, or a fetch that failed. It is held rather than discarded: a nightly job rechecks it against the same upstream sources, since GWAS Catalog and Ensembl both keep revising their own records, and something that could not be resolved last week sometimes can be today. This number is the honest other half of the variant count above — what the collector has seen but the graph cannot yet stand behind.
In two passes. A deterministic script confirms every cited PMID matches its stated author and title, that every source URL resolves, and that every MONDO/ICD-10 code denotes the entity claimed. Then a separate reviewer that did not write the page re-checks the claims against primary sources for meaning, framing and missing caveats. Both have caught real errors, which is why both exist.
Because a dataset that only shows its strengths is harder to trust, not easier. Where we could not source a figure, where an identifier does not exist yet, or where a guideline is behind a paywall, we record it and expose it through the API alongside the content. You should be able to see the edges of what we verified.
Both, and they are different things. The list endpoints serve the whole live dataset a page at a time, free and unauthenticated — 100 rows a page, with a next link, so walking the lot needs nothing from us. A Gene Knowledge Graph Snapshot is the other case: the same dataset frozen on one date, hashed, and kept at a permanent address, which is what a citation needs and what a live API can never be. Snapshots are the one paid thing here — see the section above.
One gzip-compressed UTF-8 JSON file holding every published variant, condition, drug, chemosensory group and topic as they stood on that date — with the per-genotype readings, the MONDO/OMIM/ICD-10 codes, every cited source and the date each page was last fact-checked. Plus the public record page carrying its SHA-256, which is what a reviewer opens to confirm your citation. It is a single object rather than NDJSON, so one json.load reads it, and the field names match the live API.
Because what you are paying for is not the data — it is the freezing. Keeping a dated, hashed copy of every version addressable forever is work and storage that the live API does not need, and the live API is what you should use if today’s numbers are what you want. Nothing that is free today becomes paid later; the API, every page and the bulk response stay open, and the CC BY 4.0 licence on the structured fields is unchanged whether you take it from the API or from a snapshot.
An account, yes — the file is unlocked for an account, so a payment made without one would have nothing to unlock. No expiry, no renewal, no limit on how often you download it. Charging again for data that cannot change would be renting it, not selling it. A new version is frozen every Monday, and each is a separate purchase — but almost nobody needs more than one. You buy the version your work cites; it stays valid for as long as that work does, because a dated version does not go stale.
Licensed CC BY 4.0. Reuse the structured data, including commercially, with attribution to MyGeneLog™. Written sections and entity links live on the pages. Full API docs →