Drop a MyHeritage / 23andMe / AncestryDNA / Living DNA / FTDNA / VCF export. Get a health · pharmaco · ancestry · polygenic-risk report in ~3 seconds. Your file never leaves your browser.
⚠️ Educational tool — not a medical device. DNAI does not diagnose, does not replace a medical consultation or an accredited clinical genetic test, and does not prescribe anything. Every result is presented descriptively and cited from the public literature. See/legal/termsand/legal/notice.
Consumer DNA labs hand you a pie chart of your "ethnicity", three folklore traits (bitter taste, earwax, freckles), and then they stop. Your .zip sits in an email, unused — while the same ~700,000 SNPs they read could tell you a lot more descriptively, if someone cross-referenced them with the public clinical literature.
That's all DNAI does. It opens the file you already have, annotates it against the same public databases clinicians cite (ClinVar, CPIC, DPWG, PGS Catalog, GWAS Catalog), and hands you back an editorial report — on your machine, in your RAM, never on a server.
No account. No upload. No "premium unlock for your own data". Just your file, opened properly.
| Chapter | Source | What it does |
|---|---|---|
| Health | ClinVar (NCBI) | Pathogenic / likely-pathogenic variants, filtered by review status |
| Pharmacogenomics | CPIC + DPWG | Documented drug sensitivity across ~100 molecules |
| Ancestry | AIM panel + world map | Projection on ancestry-informative markers, 5 continental groups |
| Lineages | Y-DNA + mtDNA | Paternal & maternal haplogroups + Neanderthal fragment |
| Polygenic risk | PGS Catalog | Percentile for T2 diabetes, cardiovascular, LDL, BMI… |
| Traits | GWAS Catalog + SNPedia | Caffeine, lactose, alcohol, bitter taste, eye color, etc. |
| Homozygosity | ROH segments | Inbreeding coefficient estimate |
| Animated narrative | Three.js | 3D genome fly-through, chapter by chapter |
| PDF export | — | A4 clinical-style report with cover, sections, legal block |
Not a promise — an architecture.
- No upload. Parsing & annotation run in a Web Worker, in browser memory.
- No persistence. The file is freed when you close the tab. Only your disclaimer acceptance is kept (
localStorage). - Works offline. Turn off Wi-Fi after the page loads — the whole pipeline keeps running.
- No genetic telemetry. We embed Vercel Analytics (page views, country, referrer). It never sees your file's content.
You can audit this yourself: open DevTools → Network, run an analysis, watch zero requests go out.
| Source | Typical chip | Build |
|---|---|---|
| MyHeritage | ~700k SNPs | GRCh37 |
| 23andMe | ~600k SNPs | GRCh37 |
| AncestryDNA | ~700k SNPs | GRCh37 |
| Living DNA | ~650k SNPs | GRCh37 |
| FamilyTreeDNA | ~700k SNPs | GRCh37 |
| VCF (WGS) | millions of variants | GRCh37 / GRCh38 |
The report adapts its copy to the detected source and the number of positions read.
DNA file (.csv / .zip / .vcf)
│
▼
┌──────────────────────┐
│ Web Worker │
│ ─ fflate (unzip) │
│ ─ streaming parser │
└──────────┬───────────┘
▼
┌──────────────────────┐ ┌─────────────────────────┐
│ Annotation pipeline │ ← │ public/data/*.json(.gz)│
│ ─ ClinVar slim │ │ Pre-built from public │
│ ─ CPIC / DPWG │ │ sources in scripts/ │
│ ─ PGS scoring │ │ (ClinVar, CPIC, PGS…) │
│ ─ AIM projection │ └─────────────────────────┘
│ ─ Y/mt haplogroups │
└──────────┬───────────┘
▼
Report UI · 3D narrative · PDF export
Everything above the arrows is your data. Everything to the right ships with the app.
pnpm install
pnpm dev # http://localhost:3000
pnpm test # Vitest unit tests
pnpm build # static export → out/Pure static export: deploys on Vercel, Cloudflare Pages, Netlify, GitHub Pages, or any S3 bucket. No server, no cron, no database.
To run the end-to-end smoke test against a real DNA file:
DNAI_REAL_FILE=/path/to/my-file.zip pnpm testUses pnpm, not npm. Adding a dep?
pnpm add <pkg>.
- Next.js 16 App Router, static export
- React 19 + TypeScript strict
- Tailwind CSS 3 with CSS variables (Clinique paper palette)
- Zustand for state (analysis, consent, unlock)
- fflate for in-browser zip decompression
- D3 + Three.js (
@react-three/fiber,@react-three/drei) for visualizations - Vitest for tests
- Biome for lint & format
Repo layout (high level):
app/ Next.js App Router pages (/, /story, /report, /legal/*)
components/ React components — sections/, story/, viz/
lib/
annotation/ Pipeline modules — clinvar, pharma, prs, ancestry, haplogroups
parsing/ Raw-file parsers (23andMe, MyHeritage, AncestryDNA, VCF…)
store/ Zustand stores
public/data/ Pre-built JSON annotation databases
scripts/ Scripts that rebuild public/data/ from public sources
Annotation databases are pre-built in scripts/ from the official public sources:
- ClinVar (NCBI) — clinical significance of variants
- CPIC — Clinical Pharmacogenetics Implementation Consortium
- DPWG — Dutch Pharmacogenetics Working Group
- PGS Catalog (EBI / NHGRI) — polygenic score models
- GWAS Catalog + SNPedia (hand-curated) — traits & phenotypes
Generated JSON files live in public/data/.
Honest about what DNA chips can and can't do:
- Consumer chips — up to ~40% false-positive rate on rare variants per the literature. Any significant result must be confirmed by an accredited lab before any clinical decision.
- Structural variants, CNVs, long indels — not covered by chip arrays. For a WGS VCF, only biallelic SNVs with
PASSare annotated today. - Ancestry — compact illustrative panel (5 continental groups), not a fine sub-continental ancestry test.
- No diagnosis — DNAI is neither a medical device nor a clinical genetics laboratory.
If you want a diagnosis, see a geneticist. If you want a second look at the file sitting in your drawer, use DNAI.
- Genes for Good & Nebula CSV parsers
- Extended PGS panel (Alzheimer, osteoporosis, autoimmune)
- Carrier-status report (recessive variants — CFTR, HEXA, etc.)
- WGS VCF: multi-allelic + indel annotation
- Sub-continental ancestry (K≥12 panel)
- Imputation for missing positions (BRCA1/2 high-impact SNVs)
- Full i18n (currently: French + English copy)
Issues and PRs welcome. Given the medical context, any added clinical data must:
- Cite a traceable public source (DOI, official URL).
- Be phrased descriptively ("the literature associates…", "CPIC describes…"), never prescriptively ("avoid", "recommended").
- Pass
pnpm testandpnpm build.
We explicitly don't accept contributions that turn DNAI into a diagnostic tool or add prescriptive medical advice. That line is what keeps the project safe, legal, and useful.
MIT — free to use, modify, and redistribute.
DNAI stands on the shoulders of the public-science community: NCBI / ClinVar, CPIC, DPWG, EBI / PGS Catalog, GWAS Catalog, and SNPedia contributors. None of this project would be possible without their decades of open curation. Thank you.
Built for the file already in your drawer.
