How the report is built.

01Where the data comes from

The reports are built on the NHGRI-EBI GWAS Catalog, the public, curated collection of genome-wide association studies maintained by EMBL-EBI. We use release 2026-06-01 (dbSNP 156, genome build GRCh38.p14). A scheduled workflow checks for new releases, rebuilds our catalogs and brings them into the generator after review, so the release in use can lag the newest one by a few weeks. This page shows the release currently in use.

02Which associations are kept

From more than a million catalog rows we keep the associations that meet all of these rules:

• Genome-wide significance: a reported p-value below 5 × 10⁻⁸.
• A reported effect size: an odds ratio or a beta coefficient.
• A plausible confidence interval: rows whose interval is implausibly wide (more than a factor of 3) are discarded.
• A trait mapped to a disease in the MONDO ontology. Those rows form the health catalog: 23,220 associations, 1,114 conditions and traits, 1,459 studies. Nutrition-related traits (vitamins, minerals, trace elements, food sensitivities) form a separate nutrition catalog: 10,443 associations, 292 traits, 157 studies.

03How your file is read

The generator accepts the raw-data export of 23andMe, AncestryDNA, MyHeritage, FamilyTreeDNA, Living DNA and WeGene (.txt, .csv or .zip). It reads the file in your browser's memory, takes each line's rsID and genotype, and ignores no-calls. Archives are unpacked on your device. Nothing is sent to a server.

04How a row is matched

A variant appears in your report when two things are true: your file contains that rsID, and your genotype carries the risk allele reported by the study. At present the generator scores homozygous calls only (AA, CC, GG or TT); a heterozygous call is left out rather than scored as half an effect. Variants for which you do not carry the risk allele are not listed.

05How the effect per variant is computed

Several studies often report the same variant. Their effects are averaged within each study, then across studies, giving one effect per variant. It is printed in the 'Effect' column next to the PubMed IDs it comes from.

06How the verdict per condition is computed

For conditions measured by odds ratios, the variant effects are multiplied; for traits measured by beta coefficients, they are added. A product above 1 (or a sum above 0) prints as higher risk, below as lower risk, and exactly 1 (or no matched variant) as average or undetermined. This is a deliberately simple and transparent aggregation, not a clinical polygenic risk score. In particular, it:

• does not weight the number of risk-allele copies: a scored variant counts once;
• does not correct for correlated variants (linkage disequilibrium), so neighbouring variants from the same signal can count more than once;
• does not use a baseline population risk, so 'higher risk' says nothing about absolute probability;
• treats every study equally, regardless of sample size or ancestry.

07Categories

Conditions are grouped into 17 categories through the catalog's EFO and MONDO ontology mapping. The complete health report contains all 17; each category is also available as its own PDF.

08What is deliberately excluded

The report does not interpret:

• pharmacogenomics (drug response);
• carrier status or rare pathogenic variants, including clinically actionable variants such as those in BRCA1 and BRCA2. Consumer arrays and GWAS associations are the wrong tool for them;
• ancestry, relatives or appearance traits;
• candidate-gene variants that never reached genome-wide significance, such as the common MTHFR, COMT or CYP1A2 variants discussed online.

09Known limits

Read the report with these in mind:

• A consumer array measures a few hundred thousand positions. A typical file contains only a fraction of the catalog's lead variants, often around a quarter; conditions whose variants are missing from your file show as undetermined.
• Most GWAS participants are of European ancestry; effects transfer imperfectly to other ancestries.
• An association is a statistical signal in a study population, not a cause and not a diagnosis. Effects below about 1.3 are weak.
• Providers use different chips and chip versions; two files from two providers can give different reports for the same person.
• The report is informational and not a substitute for genetic counselling or medical advice.

10Versioning

This page shows the catalog release the generator currently uses. Reports generated from different releases can differ as studies are added or re-curated; keep the generation date with your PDF if you want to compare later.

11Privacy architecture

Analysis and PDF rendering run entirely in your browser: Rust compiled to WebAssembly, with the PDF typeset by Typst. The only requests during generation reserve and complete a credit and carry the report type, never your file. The details, and a way to verify it yourself, are on the 'How we protect your DNA' page.

12Citation and corrections

If you use the catalog or the reports in your own work, cite the GWAS Catalog. If you spot an error on this page or in a report's method, write to support@full-human.com.

Cerezo M. et al. The NHGRI-EBI GWAS Catalog: standards for reusability, sustainability and diversity. Nucleic Acids Research, 2025. doi:10.1093/nar/gkae1070

Reports are informational and are not medical advice.