The genetics reference people and AI can cite.

What we build

Genetics information online is split in two. On one side sit the research databases — precise, authoritative, and impenetrable unless you already have the training. On the other sit consumer sites that are readable but thin, hedged, and often quietly wrong. Almost nobody publishes work that is rigorous and readable, and almost nobody publishes it under a license that lets others build on it.

That gap is the whole reason MyGeneLog™ exists. We are working toward detailed, carefully-sourced coverage of four things and, above all, how they connect:

• Variants — what specific, well-studied positions in the genome actually mean, genotype by genotype, with the research cited.
• Conditions — what a disease is, who it affects, how common it really is, and which variants have a genuine, replicated link to it.
• Drug response — where pharmacogenomic research shows a person's genetics can change how a medicine behaves, described honestly and never as a prescription.
• Smell and taste — the rare corner of genetics where a variant does not nudge a probability but changes what a person actually perceives. Two people meet the same molecule and one of them smells nothing at all.
• Topics — the one axis that starts outside genetics and works inwards, because a reader arrives with a word rather than an rsID. Each topic shows which of our variants turn up in the same papers, and says plainly that a shared paper is not a cause.

Each of those already exists somewhere on its own. Variants sit in one database, diseases in another, prescribing guidelines in a third, the chemical senses in the literature of a fourth. They have never been introduced to each other.

The join is the product. A position in your file is not information until it is connected to something — and connected differently, the same position becomes a different kind of answer.

Those joins have a name and a number: the Gene Knowledge Graph — 339,642 edges across 257,364 nodes, every one of them labelled with the evidence behind it. A link a person curated from a cited source is not the same claim as two things turning up in one paper, and this graph never adds the two together. A knowledge graph nobody can count is a slogan, so the number goes with the name every time it appears here.

So a person put on warfarin can reach the exact position their own file contains. A person who tastes soap in cilantro can find out why. A person who cannot digest lactose can see that it leads onward to calcium and vitamin D. None of that requires new science — all of it is already published. It requires the connections to be drawn, and drawn carefully enough that a clinician finds nothing to correct.

Connected, it stops being data and starts being about you.

How we write

Every page is written from freely-available published research, in our own words, and cited so you can check us. We say plainly when evidence is strong, when it's weak, and when a popular claim on the internet simply isn't supported — because a resource that only ever confirms what people hope to hear is worthless. Where a number is uncertain, we give the range rather than a tidy figure that looks more confident than the science is.

And we publish it openly. Every page is licensed CC BY-NC 4.0, and the variant dataset is available through a free public API with no key and no signup, so researchers, students, patient groups and clinicians can all build on it with attribution. Commercial use — inside a product, inside a company, or to train a commercial AI system — is licensed, and that is how the work is paid for.

How this is built

The people above work with Claude Code, an AI coding agent, the way they would work with a colleague: it writes and reviews, a person decides. The collectors behind the catalogue, the fact-checking scripts, the Korean translation pipeline and most of these pages were built that way — in the open, one reviewed change at a time.

It does not run the site. Twenty-two recurring jobs keep this place moving — pulling from the GWAS Catalog, re-checking citations against Europe PMC, sealing a dated Gene Knowledge Graph Snapshot every Monday — and not one of them calls a language model. They are deterministic scripts with fixed quality gates, and anything that fails a gate is refused rather than published. A page written by a machine and checked by a machine is a page nobody has checked.

So the division of labour is plain. The agent does what it is good at: writing code, catching its own mistakes, and doing the unglamorous parts at a pace a person cannot. What it does not do is decide. An operator chooses what is published and what is held back; reviewers read both editions, English and Korean, and read the science. Both have overruled it, including on this page.

Where this is going

This is not a finished catalog we're maintaining — it's one we intend to keep growing, aggressively and without an end date.

• Variants, updated continuously. New findings are pulled from the public GWAS Catalog every twenty minutes and reviewed, and installed copies of the app pick up additions automatically — usually within 24 hours. The dataset you see today is smaller than the one you'll see next month, and that will keep being true.
• Conditions, in real depth. We're expanding coverage of human disease with the detail these pages deserve: what the condition actually is, the symptoms it produces, what causes it, how common it genuinely is and in whom, how it's diagnosed, and exactly which variants have a replicated link to it — never a vague gesture at "genetic factors."
• Drug response, as far as the evidence allows. Pharmacogenomics is where genetics is already changing medicine, and it's badly explained almost everywhere. We're building out gene–drug coverage as thoroughly and precisely as published guidelines support — describing what the research shows, citing it, and stopping exactly where the evidence stops.

The standard we hold ourselves to is simple: a specialist should find our page correct, and their patient should find it readable. Anything that fails either test isn't finished yet.

The app

MyGeneLog™ is also a free Windows desktop app that reads a raw genome file — a VCF, or a raw data export from 23andMe or AncestryDNA — and matches it against this curated set of findings. Your file is parsed in memory on your own computer. Nothing is uploaded, nothing is stored on a server, and there is no account to create. Privacy here isn't a policy promise; it's an architectural fact.

What we are not

MyGeneLog™ does not perform genetic testing, and it is not affiliated with 23andMe, AncestryDNA, or any clinical testing service. Nothing here diagnoses a condition, predicts what will happen to you, or recommends a treatment. Genetics shifts probabilities; it rarely decides outcomes. For what a result means for you specifically, talk to a doctor or a licensed genetic counselor — and bring the page with you.

MyGeneLog™ is developed and maintained by narubox™. The app is actively maintained — if something goes wrong, or a citation here is wrong, tell us and we'll fix it and ship an update. We'd rather fix it than defend it.