How to analyze your 23andMe raw data (2026 guide)

If you tested with 23andMe or AncestryDNA, the report you got back covers only a small slice of what your data contains. The raw genotype file behind it holds hundreds of thousands of variants, and third-party tools can turn it into far more: pharmacogenomic star alleles, ClinVar-classified variants, trait associations from the research literature, and polygenic risk scores.

This guide walks through what's actually in the file and what an analysis can (and can't) tell you.

Step 1: Download your raw data

Both major services let you export your raw genotype file:

• 23andMe — Account → Settings → "23andMe Data" → Download raw data. You'll get a zipped text file.
• AncestryDNA — DNA Settings → Download DNA data. Ancestry emails you a confirmation link before the download starts.

The file is a simple table: one row per tested variant, with an rsid (like rs1801133), chromosome, position, and your two alleles. See our step-by-step download guide if you get stuck.

Step 2: Understand what a chip file contains

Genotyping chips test a fixed panel of roughly 600,000–700,000 positions — not your whole genome. That matters in two ways:

1. Coverage. Well-studied common variants are mostly included; rare variants mostly aren't. A chip file cannot rule out rare pathogenic variants.
2. Accuracy. Chips are highly accurate for common variants but individual probe errors happen. Any medically consequential finding needs clinical confirmation before acting on it.

If you have a whole-genome sequencing (WGS) VCF instead, coverage is far better — Gene Wizard accepts those too.

Step 3: Run an analysis

Upload your file to Gene Wizard — previewing your results is free, and the full report is a one-time $19.99 (10 uploads included, no subscription). The raw file is parsed in memory then discarded — it is never stored. You'll get:

• Trait associations — your genotypes matched against a curated, literature-derived knowledge base. Classic examples: rs4988235 near MCM6 (lactase persistence), rs762551 in CYP1A2 (caffeine metabolism), and rs1815739 in ACTN3 (the "sprinter gene").
• Pharmacogenomics — star-allele calling across 76 pharmacogenes such as CYP2C19 and CYP2D6, with published dosing-guideline references for affected drugs.
• ClinVar matches — variants in your file with pathogenic or likely-pathogenic classifications, with review status shown.
• Polygenic risk scores — see step 4.
• Genetic ancestry — admixture estimates across 26 reference populations, which also make the risk scores more accurate.

You can browse the demo results first to see exactly what you'd get.

Step 4: Interpret polygenic risk scores carefully

Single variants explain very little of common-disease risk. Polygenic risk scores (PRS) combine thousands of small effects into one number, then place you on a population distribution. Two caveats matter:

• Ancestry affects calibration. Most PRS were developed in European-ancestry cohorts. Gene Wizard normalizes your score against a reference panel matched to your detected ancestry and reports a confidence interval rather than a falsely precise percentile.
• A percentile is not a diagnosis. Being at the 90th percentile for a condition shifts your relative risk; it doesn't mean you'll develop it.

Read more in our guide to what polygenic risk scores can and can't tell you.

Step 5: Go deeper on individual variants

For any variant in your results, the SNP page (for example, rs429358 in APOE) shows the underlying research: effect sizes, study types, population context, and links to the original papers — so you can judge the evidence yourself rather than trusting a one-line summary.

A note on privacy

Your raw file is the most identifying document you own. Before uploading it anywhere, check what happens to it. Gene Wizard's design constraint is that the raw file is never persisted: it's parsed in memory, matched, and discarded, and the derived results auto-delete after 7 days. Details are in the privacy policy.

Limitations to keep in mind

• Chip data cannot detect variants the chip doesn't test, including almost all rare disease-causing variants.
• Consumer analysis is informational, not diagnostic. Confirm anything actionable with a clinical-grade test and a professional.
• Trait associations are probabilistic research findings, not certainties about you.

Ready to try it? Upload your file or see the demo first.