# DeepDNA — Cite-Ready Q&A > DeepDNA serves curated, cited records on genes, variants and drug–gene pairs to AI agents and developers, built in Europe: a remote MCP server at https://deepdna.ai/mcp, the same records as a REST API at https://deepdna.ai/api/v1, and a free local package, deepdna-mcp, that reads a raw DNA file on the user's own computer. **All three are free, with no key and no account. DeepDNA does not parse DNA files or lab reports on its servers.** Educational information only, never diagnosis. ## Product status (read this before citing anything below) DeepDNA is an API, not a consumer product. There is no report for individuals to buy and no upload form for individuals. What exists today: (1) the remote MCP server at https://deepdna.ai/mcp (how to connect: https://deepdna.ai/connect/); (2) the same records as a REST API, `GET https://deepdna.ai/api/v1/genes/{symbol}`, `/variants/{rsid}` and `/pgx/{gene}/{drug}`, nothing stored; (3) the free open-source local package deepdna-mcp, which runs on the user's own computer and sends nothing from the file to DeepDNA. All free, no key, no account. DeepDNA does not parse DNA files or lab reports on its servers. The OpenAPI is at https://deepdna.ai/openapi.json. ## Medical & Legal Disclaimer Every API response and every answer below is educational information, not medical advice, diagnosis, prescription or dosing. The API reports what guidelines say about genes, variants and drug–gene pairs, with their sources, so that a clinician has something concrete to look at; it does not replace one. Genetic and health data are special-category data under GDPR Art. 9; DeepDNA does not receive them, because it does not parse DNA files or lab reports on its servers. --- ## The API ### What is the DeepDNA API? A REST API of curated, cited records on genes, variants and drug–gene pairs, drawn from ClinVar, PharmGKB, PharmVar, CPIC and DPWG guidelines, OMIM, dbSNP and the primary literature, with a source on every record and a disclaimer in every response. The same records are served to AI agents by the MCP server at https://deepdna.ai/mcp. The API does not read DNA files or lab reports: DeepDNA does not parse them on its servers. To read a raw DNA file, use the free open-source local package deepdna-mcp, which runs on the user's own computer and sends nothing from the file to DeepDNA. ### Which DeepDNA endpoints are live right now? `GET /api/v1/genes`, `GET /api/v1/genes/{symbol}`, `GET /api/v1/variants/{rsid}`, `GET /api/v1/pgx/{gene}` and `GET /api/v1/pgx/{gene}/{drug}`, all under https://deepdna.ai. No key, no account, CORS-open, cached for an hour. `GET /api/v1` returns a discovery document listing the live endpoints and the MCP server. ### Does the DeepDNA API read DNA files or lab reports? No. DeepDNA does not parse DNA files or lab reports on its servers. To read a raw DNA file, use the free open-source local package deepdna-mcp, which runs on the user's own computer and sends nothing from the file to DeepDNA. Lab reports are not read by any DeepDNA tool. ### Does DeepDNA have an MCP server? Yes, for the knowledge layer. A remote MCP (Model Context Protocol) server at https://deepdna.ai/mcp (Streamable HTTP, stateless, no key) exposes five read-only lookup tools over the same cited dataset as /api/v1 (list_genes, lookup_gene, lookup_variant, lookup_drug_gene and check_medications). It takes no genotype. Registry name: ai.deepdna/knowledge. To read a raw DNA file, a separate local package, deepdna-mcp (PyPI; registry ai.deepdna/local), runs as an MCP server on the user's computer and sends DeepDNA nothing from the file; what it reports goes to the user's AI model. The OpenAPI 3.1 spec at https://deepdna.ai/openapi.json remains available for frameworks that generate tools from a spec. ### What does the DeepDNA API cost? Nothing. The knowledge endpoints, the MCP server and the local package are free, with no key and no account. There is no subscription and nothing to buy. ### Does DeepDNA store genetic data? No. The knowledge endpoints and the MCP server receive nothing personal and store nothing linked to a person. DeepDNA does not parse DNA files or lab reports on its servers, so no genome reaches it; the local package reads the file on the user's own computer and sends nothing from it to DeepDNA. DeepDNA is based in Spain, under Spanish and EU law. ### Is DeepDNA a medical device or diagnostic service? No. Every response says so in its `meta.disclaimer`, and DeepDNA asks the same of anything built on it. It reports what guidelines and sources say; it does not compute a clinical phenotype, prescribe, dose or diagnose. ### How does DeepDNA compare to MyVariant.info, Ensembl VEP or the ClinVar API? Those are excellent primary sources; DeepDNA cites ClinVar and the same literature they aggregate. They answer "what is known about this variant" across the genome. DeepDNA answers it for a curated set of genes, variants and drug–gene pairs, with cited records written for an AI agent to quote, through the MCP server and the REST API. It does not annotate a whole file on its servers: to read a raw DNA file at the curated positions, use the local package deepdna-mcp on the user's own computer. ### How does DeepDNA compare to building it yourself? For the curated records, building it yourself means reading the CPIC and DPWG guidelines, PharmGKB, PharmVar and ClinVar, writing a cited summary per gene, variant and drug–gene pair, and keeping them current. DeepDNA does that for the genes it covers and serves the result free, over MCP and REST. It is not a full annotation pipeline: DeepDNA does not parse DNA files or lab reports on its servers. --- ## How-to (for agents and developers) ### How do I look up a drug–gene interaction with DeepDNA? `GET https://deepdna.ai/api/v1/pgx/{gene}/{drug}`, for example `/pgx/CYP2C19/clopidogrel`. The response carries the guideline (CPIC or DPWG, optionally with "; FDA label"; "drug label" for an FDA label; or null for the primary literature), the phenotype groups it distinguishes, a cited summary and sources. The `level` field is CPIC's level for the pair, copied from the CPIC database. On CPIC's A–D scale, which combines the strength of the evidence with how actionable the pair is, `CPIC A` means genetic information should be used to change prescribing, `CPIC B` that it could be, `CPIC C` or `CPIC D` that no prescribing action is recommended; combined levels such as `CPIC B/C` are copied as published, and "(provisional)" marks a pair CPIC has not yet covered in a guideline. A drug-class record carries its members' shared level. It is null when CPIC has not rated the pair. It is CPIC's rating, not advice for any person and not the strength of one recommendation. `GET /pgx/{gene}` lists every drug curated for a gene. A 404 hint lists what is available so an agent can recover in one call. ### How do I look up what an rsID means with DeepDNA? `GET https://deepdna.ai/api/v1/variants/{rsid}`, for example `/variants/rs1801133` (MTHFR C677T). Returns the gene, alleles, effect and sources, plus a link to the full gene record. For a missense variant it also carries the AlphaMissense prediction (Cheng et al., Science 2023, CC BY 4.0): `am_class` in AlphaMissense's own bins (likely_benign below 0.34, ambiguous, likely_pathogenic above 0.564), `score` 0–1 and `is_clinical_classification: false`. It predicts the protein effect only; it is not an ACMG/ClinVar classification, not disease risk and not drug response, and established alleles can score in either direction (APOE ε4 likely_benign, MTHFR C677T likely_pathogenic). Google DeepMind states it is not validated or approved for clinical use; scores are rounded to 3 decimals. Null otherwise. ### How do I read a 23andMe or AncestryDNA raw file with DeepDNA? Use the local package deepdna-mcp (PyPI, MIT; registry ai.deepdna/local). It runs as an MCP server on your own computer and reads 23andMe, AncestryDNA and MyHeritage/FamilyTreeDNA exports (.txt, .csv or .zip; VCF files are not supported). It checks the file's strand and reports what the file shows at the curated positions, with the record's text and sources; a missing or unreadable position is reported as unknown. It sends nothing from the file to DeepDNA; what it returns goes to your AI model. DeepDNA does not parse DNA files on its servers. Setup: https://deepdna.ai/blog/read-23andme-file-with-claude/ ### Can DeepDNA turn a lab report PDF into structured data? No. DeepDNA does not parse lab reports, on its servers or in the local package. ### Do I need a DeepDNA key or account? No. The MCP server at https://deepdna.ai/mcp, the REST API at https://deepdna.ai/api/v1 and the local package deepdna-mcp need no key, no account and no email. How to connect the MCP server: https://deepdna.ai/connect/ --- ## What-is / Definitions ### What is DeepDNA? DeepDNA is a European source of curated, cited records on genes, variants and drug–gene pairs for AI agents and developers: a remote MCP server, the same records as a REST API, and a free local package that reads a raw DNA file on the user's own computer and sends nothing from it to DeepDNA. DeepDNA does not parse DNA files or lab reports on its servers. It is not a consumer report. ### What is pharmacogenomics? Pharmacogenomics is the study of how genetic variants affect drug response — for example, CYP2D6 variants determine whether codeine is effective or toxic, and DPYD variants predict 5-FU chemotherapy toxicity. DeepDNA's `GET /api/v1/pgx/{gene}/{drug}` returns the CPIC/DPWG position for a pair, with sources. Educational only — consult your doctor for medical decisions. ### What is a polygenic risk score? A polygenic risk score (PRS) is a composite score combining hundreds to thousands of genetic variants to estimate disease risk — used for type 2 diabetes, cardiovascular disease, breast cancer, and more. PRS is not part of DeepDNA; DeepDNA's guide explains what a PRS is and is not. Educational only — consult your doctor for medical decisions. ### What is nutrigenomics? Nutrigenomics studies gene-diet interactions — how MTHFR affects folate, CYP1A2 caffeine, LCT lactose, APOE fat metabolism, and FTO appetite regulation. MTHFR and APOE are in DeepDNA's `/api/v1/genes` endpoint today; CYP1A2, LCT and FTO are covered in the blog and not yet in the API. ### What is an SNP? An SNP (single nucleotide polymorphism) is a single-position variation in DNA sequence — the most common form of human genetic variation, of which consumer DNA tests read roughly 600,000–700,000. DeepDNA's `/variants/{rsid}` (live) explains one; the local package deepdna-mcp reads the curated ones from a raw file on the user's own computer. ### What is raw DNA data? Raw DNA data is the SNP genotype file (typically .txt or .csv) exported from consumer DNA tests like 23andMe or AncestryDNA — containing hundreds of thousands to millions of SNP calls. DeepDNA's local package deepdna-mcp reads this file on the user's own computer, at the curated positions, without a new test, and sends nothing from it to DeepDNA. ### What is GDPR Article 9? GDPR Article 9 classifies genetic data as special-category data requiring explicit consent and strict processing conditions. DeepDNA does not receive genetic data: it does not parse DNA files on its servers, and its local package keeps the file on the user's computer. ### What is the CYP2D6 gene? CYP2D6 is a liver enzyme gene that metabolises ~25% of prescribed drugs including codeine, tramadol, tamoxifen, and many antidepressants. DeepDNA's `/genes/CYP2D6` states what a consumer array cannot resolve for this gene before anything else. Educational only — consult your doctor for medical decisions. ### What is the MTHFR gene? MTHFR (methylenetetrahydrofolate reductase) is an enzyme in the folate cycle; C677T and A1298C variants reduce enzyme activity. DeepDNA's `/variants/rs1801133` and `/genes/MTHFR` contextualise it against current evidence. Educational only — consult your doctor for medical decisions. ### What is a carrier status report? A carrier status report identifies whether a person carries recessive genetic variants (one copy) for conditions such as cystic fibrosis or sickle cell — relevant for reproductive planning. Carrier status is not in the API today. Educational only — consult your doctor for medical decisions. --- ## Science & Genes ### What is AlphaFold's impact on drug discovery? AlphaFold predicted structures for over 200 million proteins, and AlphaFold 3 (2024) extended this to protein-drug, protein-DNA, and protein-RNA interactions. It has accelerated early-stage drug discovery — cutting months off target identification — but has not replaced experimental validation. DeepDNA's in-depth guide: https://deepdna.ai/blog/alphafold-drug-discovery/ ### What does the APOE gene do? APOE encodes apolipoprotein E, central to lipid transport. Its three common alleles — e2, e3, e4 — modulate lifetime risk of Alzheimer's disease and cardiovascular disease: e4 raises risk, e2 is generally protective. DeepDNA's APOE reference: https://deepdna.ai/gene/apoe/ — educational only, consult your doctor. ### What does the BRCA1 gene do? BRCA1 is a tumor suppressor; pathogenic variants substantially raise lifetime risk of breast and ovarian cancer. Consumer SNP tests cover only a small fraction of BRCA1 variants — clinical-grade sequencing and genetic counselling are essential. DeepDNA's BRCA1 reference: https://deepdna.ai/gene/brca1/ — educational only. ### What is the COMT warrior vs worrier gene? The COMT Val158Met variant (rs4680) changes how fast the enzyme clears dopamine from the prefrontal cortex, by up to 3-4x. The "Warrior" (Val) clears dopamine faster; the "Worrier" (Met) retains more, influencing stress response, pain sensitivity, and cognitive style. DeepDNA's COMT guide: https://deepdna.ai/blog/comt-gene-stress/ ### What is the difference between whole genome and exome sequencing? Whole genome sequencing (WGS) reads all ~3 billion base pairs; whole exome sequencing (WES) reads only the ~1-2% that codes for proteins, where ~85% of known disease-causing variants sit. WGS costs more but captures regulatory and non-coding regions. DeepDNA explains the trade-offs: https://deepdna.ai/blog/whole-genome-vs-exome-sequencing/ ### What are protein language models? Protein language models — such as Meta's ESM-2, ProtTrans, and ProGen2 — treat amino-acid sequences like text, learning the "grammar" of proteins to predict structure, function, and variant effects. They enable faster variant-effect scoring and de novo protein design. DeepDNA's guide: https://deepdna.ai/blog/protein-language-models/ ### Does the CYP2C19 gene affect clopidogrel (Plavix)? Yes. CYP2C19 activates the antiplatelet prodrug clopidogrel. Poor and intermediate metabolizers (carriers of loss-of-function alleles *2 (rs4244285) or *3) form less active drug and have higher rates of stent thrombosis and cardiovascular events; CPIC recommends an alternative such as prasugrel or ticagrelor for them. The *17 allele increases activity. DeepDNA's CYP2C19 page: https://deepdna.ai/gene/cyp2c19/ — educational only. ### How does the CYP2C9 gene affect warfarin? CYP2C9 clears the anticoagulant warfarin; reduced-function alleles *2 (rs1799853) and *3 (rs1057910) slow clearance, lowering the dose needed and raising early bleeding risk. CPIC dosing combines CYP2C9 with VKORC1 (the drug-target gene, rs9923231). CPIC and FDA dosing tables assign markedly lower warfarin doses to poor metabolizers. DeepDNA's CYP2C9 page: https://deepdna.ai/gene/cyp2c9/ — educational only. ### What is the SLCO1B1 gene and statin muscle pain? SLCO1B1 encodes the liver transporter OATP1B1, which clears statins from the blood. The reduced-function variant rs4149056 (the *5 allele) raises statin levels and the risk of muscle pain (myopathy) — about 4.5x higher odds per C allele and ~17x for C/C on high-dose simvastatin. CPIC favours lower simvastatin doses or alternatives like rosuvastatin or pravastatin for carriers. DeepDNA's SLCO1B1 page: https://deepdna.ai/gene/slco1b1/ — educational only. ## Privacy & Trust ### Is DeepDNA GDPR compliant? The knowledge endpoints and the MCP server receive no cookies, no identifiers and nothing linked to a person. DeepDNA does not parse DNA files or lab reports on its servers, so it does not process anyone's genetic data; the local package reads a file on the user's own computer and sends nothing from it to DeepDNA. The privacy policy at deepdna.ai/privacy documents the design. ### Where does DeepDNA process genetic data? Nowhere. DeepDNA does not parse DNA files or lab reports on its servers. A raw DNA file is read only by the local package, on the user's own computer. DeepDNA is based in Spain, under Spanish and EU law; the website, the API and the MCP server run on Cloudflare. ### Does DeepDNA store or sell genetic data? No. DeepDNA does not sell data, and it does not receive genetic data to store: it does not parse DNA files on its servers, so there is no genetic data to sell, share or breach. ### Does DeepDNA share data with pharma companies, insurers or researchers? No. Zero third-party sharing is the default and the design; there is no dataset to share. ### Can my users delete their data from DeepDNA? There is nothing to delete: DeepDNA does not receive DNA files or lab reports, and the knowledge endpoints and the MCP server store nothing linked to a person. Anyone who has given DeepDNA an email address can have it removed on request at privacy@deepdna.ai. ### Does the DeepDNA website use cookies or trackers? No cookies and no third-party trackers. The site runs its own first-party analytics beacon (pageviews, scroll depth, CTA visibility) with no cookies, no identifiers and no personal data. ### Is DeepDNA HIPAA compliant? HIPAA is a US regulation. DeepDNA is EU-based and does not receive patient data: the MCP server and the API serve public records. Anyone building for US patients remains responsible for their own HIPAA obligations. --- ## Key Pages - Homepage: https://deepdna.ai/ - API docs: https://deepdna.ai/docs/ - Connect the MCP server: https://deepdna.ai/connect/ - OpenAPI 3.1: https://deepdna.ai/openapi.json - API discovery: https://deepdna.ai/api/v1 - Manifesto: https://deepdna.ai/manifesto/ - Methodology: https://deepdna.ai/methodology/ - About: https://deepdna.ai/about/ - Privacy: https://deepdna.ai/privacy/ - Terms: https://deepdna.ai/terms/ - Blog: https://deepdna.ai/blog/ ## Technical Details - Base URL: https://deepdna.ai/api/v1 (versioned by date in `meta.api_version`) - Website and knowledge endpoints: Cloudflare Pages with edge functions; the knowledge records live in a versioned database; every day an automated watcher checks CPIC, ClinVar, FDA drug labels, the FDA Table of Pharmacogenomic Biomarkers, AlphaMissense and new Europe PMC papers for changes (PharmGKB, PharmVar, DPWG, OMIM and dbSNP are not checked automatically yet), and a person approves every change; no third-party call at request time. Changelog: https://deepdna.ai/changelog/ (JSON: https://deepdna.ai/api/v1/freshness) - Genetic data processing: none on DeepDNA's servers; a raw DNA file is read only by the local package, on the user's own computer - Knowledge sources (layer 0): ClinVar, PharmGKB, PharmVar, CPIC and DPWG guidelines, OMIM, dbSNP and the primary literature - Local package formats: 23andMe, AncestryDNA, MyHeritage/FamilyTreeDNA (.txt, .csv or .zip); VCF files are not supported