Ves al contingut principal

Aquest contingut encara no està disponible en català. El mostrem en anglès.

Equip DNA Info Lab · Revisió editorial · · 12 min de lectura

How to Interpret Your 23andMe Raw Data: A Practical Guide (Updated 2026)

How do you interpret 23andMe raw data? Check the file, find a variant by its rsID, read its genotype and compare it with a source using the same genome build and strand. You can do this manually or use an analysis tool. A match is information to investigate, not a diagnosis; a missing row is not a negative result.

If you've downloaded your raw DNA file from 23andMe (or AncestryDNA, or MyHeritage — the principle is the same), you have a plain text file with hundreds of thousands of rows: 23andMe's current v5 chip is a customised Illumina Global Screening Array that tests about 654,000 preselected variants plus about 50,000 custom ones. Each row is a genetic position the testing chip measured. The natural question: what can you actually do with it?

This guide follows two ways to interpret an existing file: manual lookup and automated matching against databases. Start with the checks below, then choose the method that fits your question.

First: check your file is usable

Before interpreting anything, spend one minute on the file itself. A surprising share of "my results look wrong" stories are really file problems: a truncated download, the wrong export, or a format a tool cannot read.

The fastest check is our free raw DNA file checker, which runs entirely in your browser — the file never leaves your device. It tells you which provider and format the file is, how many variant rows it contains, how many no-calls there are, and whether it looks truncated. A text editor works too: a healthy 23andMe file starts with a few header lines — in 23andMe's own example they name build 37 and the plus strand — followed by row after row of four tab-separated values.

  • Right file? You want the text file whose name begins with genome, something like genome_Firstname_Lastname_Full_*.txt, delivered zipped — not the ancestry composition PDF or a health-report summary.
  • Does it look complete? File size and marker count vary by chip and export. Check for malformed rows or an abrupt ending; download again if you suspect an incomplete export. A format check cannot certify accuracy or clinical completeness.
  • Readable format? 23andMe and AncestryDNA ship tab-separated .txt; MyHeritage ships a quoted .csv. Tools that auto-detect the format save you a conversion step.

What's actually in the file

The 23andMe raw file is a tab-separated text file with four columns:

rsidchromosomepositiongenotype
rs49882352136608646AG
rs1801133111856378--
rs60251169519049CT
rs4293581945411941CT
  • rsid — the variant's Reference SNP cluster ID, the unique identifier assigned in dbSNP (e.g. rs429358 = the APOE variant). IDs that start with i are 23andMe's internal ones and may have no rsID elsewhere.
  • chromosome — 1-22, X, Y, or MT for mitochondrial DNA: 23andMe describes your genome as 23 chromosome pairs plus the DNA in your mitochondria.
  • position — the exact base-pair position on that chromosome.
  • genotype — the two letters you have at that position (one inherited from each parent).

This is not your whole genome. The file contains the positions measured by the chip; coverage varies. An absent variant may simply not have been tested. See the file format guide for columns, no-calls and chip versions.

Before a manual lookup: five checks

  1. Find the exact rsID in your file and read the genotype from the same row. If it is missing, record it as unavailable.
  2. If the genotype is --, stop: it is a no-call, not evidence that you lack the variant.
  3. Check the reference genome in the file header and the strand used by the source. 23andMe states that it reports genotypes on the plus (+) strand of GRCh37 by default, and also on the plus strand of GRCh38 in its Browse Raw Data tool; a source written on the other strand or build can show different letters or coordinates for the same variant.
  4. Read the source record: which alleles, condition or trait, study population and evidence does it describe? A database entry is not an interpretation of your health.
  5. Keep the rsID, genotype, source and uncertainty together. Take a concerning finding to a qualified professional before making health or medication decisions.

Method 1: Look up specific variants by hand

The most direct way is to look up specific rsids you've heard about. For each variant, you can check what your letters mean in three databases:

To explore a variant, open our public rsID lookup and enter only its identifier, for example rs4988235. It queries public reference information without uploading your DNA file; it does not interpret your genotype.

Famous variants people look up

This is informative but slow — each variant takes several careful steps, as the example below shows, and you have to be careful about ancestry-specific effects and overinterpreting single SNPs.

A worked example: lactose tolerance

Say you want to know what your file says about lactose. The best-studied variant is rs4988235, in the regulatory region of the lactase gene. Here is the whole manual method, end to end:

  1. Open your raw file in a text editor and search for rs4988235. You will find a row whose last column is your genotype: AA, AG or GG.
  2. On the plus strand that 23andMe reports, dbSNP lists this position as G or A. The A allele corresponds to the T of the −13910 C/T variant on the lactase gene's own strand, which ClinVar records as associated with lactase persistence — the ability to digest lactose as an adult — and in the GWAS Catalog it is associated with higher blood levels of the lactase enzyme. GG carries no copy of it. The variant was first linked to lactase non-persistence in Finnish families (Enattah et al., 2002).
  3. Verify the association in a second source: the GWAS Catalog record for rs4988235 lists each study that reported it, and each study's page describes the population it recruited.
  4. Note the limits before concluding anything: this variant is associated with lactase persistence in Europeans, while in East African populations other variants are associated with it. A genotype also cannot tell you whether you have symptoms. It is a research association, not a diagnosis.

That is one variant, and it took four careful steps. The method is right — and it is exactly why doing this by hand for hundreds of positions is not realistic, which is where Method 2 comes in.

Method 2: Cross-reference against curated databases

Several tools take your full raw file and automatically match it against curated, peer-reviewed databases. The three databases that matter most are:

  1. GWAS Catalog → research associations. Finds which of the hundreds of variants studied for traits like type 2 diabetes, heart disease, Alzheimer's and depression are in your file, and what each study reported for them. It does not add them into a score.
  2. PharmGKB → pharmacogenomics. Tells you how your genotype may affect drug response for warfarin, statins, codeine, clopidogrel, SSRIs, metformin and others.
  3. ClinVar → clinical variants. Flags positions where you carry a variant labs have classified as pathogenic or likely pathogenic for a specific condition.

Our service matches the file against GWAS Catalog, PharmGKB and ClinVar and links the results to their sources. ClinVar classifications come from database submissions; the report is not an individual clinical review. See the 23andMe analysis page for the report sections and what you can and cannot learn for their limits.

Do you need a new test to read this file?

No new test is needed to read the rows you already have. Whole-genome sequencing generates a different dataset; it is not a third way to interpret this export or recover untested positions from it. Whether another test is appropriate for a medical question requires professional assessment.

How to read an odds ratio without fooling yourself

Most GWAS results are reported as an odds ratio (OR): the odds of the trait in carriers of a variant divided by the odds in non-carriers, as measured in one study. Three habits keep this honest:

  1. An odds ratio is not your probability. OR 1.5 does not mean "50% chance", and it does not mean your risk grew by half an absolute point. Applied to a common trait it can shift your personal probability noticeably; applied to a rare one, the same OR moves it by a fraction of a percent.
  2. Baseline matters more than the multiplier. Doubling a 1-in-10,000 risk gives 1-in-5,000 — real, but small in absolute terms. Doubling a 1-in-5 risk is a different conversation entirely.
  3. One study is a hint, not a verdict. Associations replicate or they fade. A variant with one small study behind it deserves a shrug; one confirmed across many large studies — like the APOE example below — deserves attention. And effect sizes measured mostly in European-ancestry participants may not transfer to you unchanged.

Common mistakes when interpreting raw data

  • Reading a no-call as a result. A genotype of -- (23andMe) or 0 (AncestryDNA) is a no-call: 23andMe writes -- when its algorithm cannot make a confident genotype call at that position. It does not mean you lack the variant. Draw no conclusion from it.
  • Ignoring the genome build. The positions in your file refer to a reference genome, usually GRCh37. Databases that display GRCh38 coordinates show different numbers for the same variant — match by rsID, which is stable, not by position.
  • Confusing the DNA strand. Some databases report genotypes on the opposite strand, where A↔T and C↔G are swapped. An "AA" there can be the "TT" in your file. Reputable tools handle this silently; manual lookups need the check.
  • Treating one SNP as a story. Common complex diseases are influenced by many variants together with environmental factors — NHGRI notes there can be hundreds or even thousands per disease. The interpretable signal is the aggregate — which is what polygenic scores formalise and what curated databases put in context.

A privacy checklist before you upload anywhere

Your raw file is the most identifying document you own — unlike a password, it cannot be reissued. Before uploading it to any service, ours included, four questions should have one-sentence answers:

  1. Where is it processed? Look for a jurisdiction you are comfortable with. Our servers are in Germany, under GDPR.
  2. What is stored, and for how long? We do not keep the file itself; the extracted variants and your report stay until you delete them, and inactive accounts are purged automatically.
  3. Can I export and delete everything? With us: yes, self-service, from your account page. If a service makes you email support to delete your genome, walk away.
  4. Which providers receive data? We do not sell genetic data. Hetzner hosts the service in Germany; uploads pass through Cloudflare. Optional AI explanations send selected results to Anthropic. Check the scope and retention in our privacy policy before uploading.

The big caveats

Before you go variant-hunting, three things to keep in mind:

  1. Most studies are biased toward European ancestry. As of June 2021, 86% of the samples in genomic studies, GWAS included, came from people of European descent. The effect sizes you read about were measured there, and may not apply to you with the same magnitude.
  2. A single variant rarely means much. Even the APOE ε4 allele — which is not a single variant but a combination of rs429358 and rs7412, and which this report does not derive — is published as odds ratios that vary widely with the number of copies and the population studied. One meta-analysis reports 3.2 for one copy (ε3/ε4) and 14.9 for two (ε4/ε4) in Caucasian samples, and 5.6 and 33.1 for the same two in Japanese ones. Most common diseases are polygenic: many variants together with environmental factors.
  3. This is not medical advice. Your raw data is a starting point for a conversation with a doctor, not a diagnosis. Anything that looks alarming deserves a clinical follow-up, not a panic.

Doing it the easy way

Manual lookup helps you understand individual records. For a broader view, compare the report sections and limitations on our analysis page, or check the file locally before deciding whether to upload. Automated matches still need context; they do not replace clinical confirmation.

Frequently asked questions

How long does it take to interpret 23andMe raw data?▼

There is no fixed duration. A manual lookup depends on the number of variants and how much evidence you review. Automated processing depends on file size and service load; reading and assessing the results takes additional time. Processing speed does not establish clinical validity.

Can I open my 23andMe raw data file in Excel?▼

Technically yes — it's a tab-separated text file. But it has hundreds of thousands of rows, which can make a spreadsheet slow, and spreadsheets may silently reformat values they mistake for numbers or dates. For a one-off look it's fine; for analysis use a text editor like VS Code or a script.

Is it safe to upload my raw DNA data online?▼

Read the service’s privacy policy first. We parse the raw file in memory and store extracted variants and reports with Hetzner in Germany. Uploads pass through Cloudflare. If you request an optional AI explanation, selected results are sent to Anthropic; the raw file itself is not sent to the AI. We do not sell genetic data. You can export or delete your data from your account.

What's the difference between rs429358 and rs7412?▼

These are two positions used when describing APOE alleles. Reading one row does not establish an APOE diplotype. Our report does not derive APOE ε2/ε3/ε4 status; do not assume every interpretation tool does so or that a raw-data lookup confirms a clinical result.

How do I know which chip version 23andMe used for my sample?▼

Check the file header and the information supplied with your test. If the chip version is not specified, ask the provider rather than guessing from file size. The measured marker set and no-calls determine what can be matched; a newer chip does not guarantee complete clinical coverage.

Will my report be different if I use AncestryDNA or MyHeritage data instead?▼

Yes, the matches can differ. Providers and chip versions test different sets of positions, and no-calls also affect coverage. Format compatibility does not guarantee the same GWAS, pharmacogenomic or ClinVar matches. Check the coverage shown for your actual file.

What does a genotype of "--" mean in my file?▼

It is a no-call: 23andMe's algorithm could not make a confident genotype call at that position, so the file shows "--" instead of two letters. It does not mean the variant is absent, and interpretation tools simply skip no-calls.

Which genome build is my 23andMe data on?▼

Check the reference assembly recorded in the file header rather than assuming it from the provider name. 23andMe documents plus-strand reporting and GRCh37 as the default in its raw-data browser, with GRCh38 also available there. When comparing a source, verify the rsID, alleles, assembly and strand together.

Can my raw data tell me whether I will get a disease?▼

No. Research associations are odds ratios measured across groups — they nudge probabilities, they do not issue verdicts. Even the strongest common variants change risk by factors, not certainties, and most trait risk is spread across hundreds of variants plus environment. Treat anything that worries you as a reason to talk to a doctor, not as a diagnosis.

Related reading

Ja tens el teu fitxer d'ADN brut?

Explora les coincidències, explicacions, límits i fonts disponibles per al teu fitxer.

Related articles

Aquest article té finalitats exclusivament educatives i no constitueix consell mèdic, diagnòstic ni tractament. Consulta sempre un professional sanitari per a decisions sobre la teva salut.