ClinVar 2026-07: 951 records inside our regions changed classification since 2026-05. What changed
Sample report Support Log in
aimostı Check your file

Measured on real files

What your DNA file can actually read

We ran ten real chip exports and four real 30x genome files through the same code that builds our reports, and counted what each one could read. The chips read between 20 and 37 percent of the 1,572 single positions the report scores. A complete whole-genome gVCF proved 99.0 percent of the regions the report examines. A plain VCF of the same kind of genome proved almost none of them, although it held every variant it found.

Written by · Published 2026-10-06. Last reviewed .

Chip exports

A chip export from 23andMe, AncestryDNA, MyHeritage or a similar service is a list of the positions that company's chip was built to measure, with your two letters at each. The ten files below held between 563,320 and 955,958 rows. What matters for a report is whether the positions it needs are among them, and whether each call can be read safely. A position counts here only when our reader would use it: calls on the ambiguous A/T and C/G pairs, insertions, deletions, garbled rows and no-calls do not count.

Scored positions each chip export could read, by part of the report
File Polygenic score variants
of 1,457		Medication-response positions
of 14		Single-gene health markers
of 12		Traits
of 29		Frontier and Fringe
of 60		 All scored positions
23andMe v32014 · 955,958 rows 4651192143 35%
23andMe v42020 · 594,993 rows 3441182343 27%
23andMe v52019 · 621,125 rows 45512102338 34%
AncestryDNA v22018 · 649,418 rows 4637101839 34%
AncestryDNA v22024 · 674,131 rows 4951292338 37%
MyHeritage2018 · 701,929 rows 28132715 20%
FamilyTreeDNA2020 · 600,334 rows 4689102031 34%
FamilyTreeDNA2023 · 686,734 rows 28332714 20%
Living DNA2019 · 691,603 rows 390182239 29%
Genes for Good2025 · 563,320 rows 430861733 31%

Most of the scored positions belong to the five polygenic scores, and the chips read between 281 and 495 of those 1,457. That is enough to place a score in a broad band, which is how our chip report shows it, and not enough for a precise percentile. The medication-response positions range from 1 to 12 of 14, so whether a chip can say anything about a given drug depends on which chip it was.

Two files read far less than the rest. Both come from the same kind of array, and it does not include many of the clinical positions: we checked by hand that the missing ones are absent from the files, not misread. Chip versions matter as much as brands, as the three 23andMe rows show.

What no chip should be asked

Chips measure common variants very well. They are poor at rare ones, which is where most disease-causing changes are. A 2021 study in the BMJ compared chip calls with sequencing in nearly 50,000 UK Biobank participants: for variants rarer than 1 in 100,000, only 16 percent of the chip's positive calls were confirmed. For disease-causing BRCA1 and BRCA2 variants, 4.2 percent of positive chip results were real. Our chip report therefore leaves out carrier status and rare clinical findings altogether, rather than show calls it cannot stand behind.

Whole-genome files

A sequenced genome covers far more than a chip, but what a report can prove depends on the file you download. A plain VCF lists only the places where you differ from the reference genome. A gVCF adds reference blocks, stretches where the sequencing looked and found the reference base. We cut each file to the 1,054 regions our report examines, 25,497,095 bases in all, and counted the bases each file positively confirms: a called genotype, variant or reference, at a read depth of at least 10.

How much of the report's examined regions each genome file confirms
File Bases confirmed Regions at least 90% confirmed Variants listed in these regions
A second 30x providergVCF 14.2% 0 of 1,054 50,015
Sequencing.com 30xgVCF · 2024 99.0% 1,038 of 1,054 41,952
Dante Labs 30xplain VCF · 2024 0.16% 0 of 1,054 41,164
Nebula Genomics 30xplain VCF · 2025 0.17% 0 of 1,054 43,261

The two plain VCFs listed tens of thousands of variants in these regions, as a good 30x genome should, and confirmed almost nothing else. Where they are silent, the report can only infer that you match the reference: the file cannot say whether the position matched or was never read. We label those readings as inferred, and on a plain VCF we never call a gene examined and clear. Why a missing line is not a clean result goes into the ways that inference has gone wrong on real files.

Not every gVCF is complete. The other gVCF in the table confirmed only 14.2 percent: the file simply has no record for most of the panel, because the pipeline that wrote it skipped stretches of the genome. This is why the report measures what each file covers, gene by gene, instead of assuming it from the file type.

If your provider offers a gVCF, download it. If it offers aligned reads, a BAM or CRAM, those let a gene be re-read base by base, which is what our Deep Read add-on does.

How we measured

The files are public exports from the Personal Genome Project, whose participants share their data openly. We name them only by vendor and year. The chips were read with the same reader that builds our chip report, over the 1,572 single positions it scores, leaving out the ancestry-line markers, which have no rsID and half of which exist only on a Y chromosome. The genome files were cut to our panel with bcftools and measured with the same coverage code that fills a report's coverage table. One file of each kind is one data point, so a different kit version can read more or less than the row shown. The measurement was run on 2026-10-06, and the numbers are available as JSON under CC BY 4.0.

To see what your own file holds, open it in the free file check. It runs in your browser, and nothing is uploaded.

Common questions

How much of my DNA does a 23andMe or AncestryDNA file contain?

A chip export lists a fixed set of positions the chip was built to measure: between 563,320 and 955,958 rows in the ten files we measured. Of the 1,572 single positions our report scores, those files could read between 20 and 37 percent.

Can a chip file tell me whether I am a carrier of a genetic condition?

Not reliably. Disease-causing changes are usually very rare, and chips are poor at very rare variants: in a 2021 BMJ study only 16 percent of chip calls for variants rarer than 1 in 100,000 were confirmed by sequencing. Our chip report leaves carrier status out for that reason. A sequenced genome is what can answer it.

Is a gVCF better than a plain VCF?

Both list the variants a sequenced genome found. A gVCF also records the stretches where the sequencing looked and found the reference, so it can prove a gene was examined. In our measurements a complete gVCF proved 99.0 percent of the regions our report examines; the two plain VCFs proved well under 1 percent, only the bases where they listed a variant.

See what your file supports before you pay anything.

Check your file, free See the reports · from €19

Sources

Aimosti re-analyses data you already own. It never takes a sample, sequences, tests or diagnoses. Every finding restates a published source, to discuss with a clinician where it touches health.