BRENNEN

Industries / Health / Healthcare / Medical labs / Genetic testing and diagnostics

Tests sold to consumers

Selling DNA tests straight to the public, usually as a saliva kit ordered online, with ancestry, trait and health reports sent back without a doctor ordering the test.

7

Milestones achieved, from 2007 to 2021.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 2007

    Genome scans sold online

    On 16 November 2007 deCODE genetics of Iceland opened deCODEme, a subscription service that read more than a million sites in a customer's DNA from a cheek swab and reported ancestry, traits such as hair and eye color, and variants linked to higher or lower risk of common diseases. The New York Times reported the same day that 23andMe, a Google-financed company founded by Linda Avey and Anne Wojcicki, was about to announce a similar service, and put the price of such scans at about USD 1,000.

    decode.comdericbownds.net

    Credit

    deCODE genetics; 23andMe

  2. 2009

    Two scans, different answers

    Pauline Ng, Sarah Murray, Samuel Levy and J. Craig Venter compared the disease risks that two direct-to-consumer companies reported for the same people, found differences between them, and published nine recommendations to make such predictions agree.

    LessonWithout shared rules for which variants and studies count, the same DNA can produce different risk reports.

    nature.com

    Credit

    Pauline C. Ng, Sarah S. Murray, Samuel Levy and J. Craig Venter, in Nature

  3. 2010

    Germany requires a doctor

    Germany's Gene Diagnostics Act (Gendiagnostikgesetz) took effect on 1 February 2010. It allows a diagnostic genetic test only by a physician, and a predictive test, one that forecasts a disease a person does not yet have, only by a specialist in human genetics or another physician qualified in genetic testing. A health test that a consumer orders and reads with no doctor involved has no place under that rule.

    gesetze-im-internet.degesetze-im-internet.de

    Credit

    German federal legislature

  4. 2013

    FDA halts health reports

    On 22 November 2013 the US Food and Drug Administration (FDA) told 23andMe to stop marketing its Personal Genome Service, which then offered health reports on 254 diseases and conditions, because the agency still had no assurance that the company had validated the tests, analytically or clinically, for their intended uses. The letter said a false positive on the BRCA breast and ovarian cancer report could lead to needless surgery. 23andMe stopped its health reports and sold ancestry reports and uninterpreted raw data instead, and in April 2014 the FDA called that response adequate.

    LessonA test sold as a "first step in prevention" is regulated as a medical device, whatever the seller calls it.

    bio-itworld.commddionline.com

    Credit

    US FDA

  5. 2017

    Risk reports cleared for sale

    In February 2015 the FDA let 23andMe sell a carrier test for Bloom syndrome straight to consumers. A carrier has one copy of a variant that can cause disease in a child if both parents carry it. The decision rested in part on a study in which 678 people at five US sites read sample reports for a fictional person. On 6 April 2017 it authorized 23andMe reports on genetic risk for 10 conditions, among them late-onset Alzheimer's and Parkinson's disease, and its rules now let a company that has had one such test cleared sell most further ones without another premarket review. A report on three BRCA1 and BRCA2 variants followed on 6 March 2018; its decision summary notes that more than 1,000 risk variants are known in those genes, so a negative result does not rule them out.

    accessdata.fda.govaccessdata.fda.govaccessdata.fda.govfda.gov

    Credit

    23andMe; US FDA

  6. 2019

    26 million in databases

    By the start of 2019 more than 26 million people had added their DNA to the four largest consumer ancestry and health databases, led by Ancestry and 23andMe, after a record year of sales, with tests selling for as little as USD 59. The figure is MIT Technology Review's own estimate, built from company statements, because the companies publish user numbers only now and then.

    technologyreview.com

    Credit

    MIT Technology Review

  7. 2021

    Chips miss rare variants

    Consumer tests read DNA with SNP chips, which check set positions instead of reading every letter, and many companies let customers download the raw results. In 2018 Ambry Genetics reported that 40% of the variants in consumers' raw data that were sent to it for clinical confirmation were false positives, in 49 cases. In 2021 Michael Weedon and colleagues at the University of Exeter compared chip results with sequencing for 49,908 UK Biobank participants: accuracy was above 99% for common variants, but only 16% of chip calls for very rare variants, those with a frequency below 0.001% in UK Biobank, were confirmed. For disease-causing BRCA1 and BRCA2 variants, only 4.2% of the chip's positive calls were real.

    LessonA chip result for a rare disease variant needs a clinical confirmation test before anyone acts on it.

    pmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

    Credit

    Stephany Tandy-Connor and colleagues, Ambry Genetics; Michael N. Weedon, Caroline F. Wright and colleagues, University of Exeter

Watch for

  1. Sequencing without a referral

    Labcorp's Marker by Labcorp Genetic Health Panel reads 163 genes linked to more than 100 medically actionable conditions, among them hereditary cancers and heart disease. Consumers buy it online through Labcorp OnDemand, with no doctor's referral, give blood at a Labcorp center and get results through an app, with access to licensed genetic counselors; it went on sale on 3 August 2026. Labcorp says nearly one in six adults who are tested carry a variant linked to a serious, medically actionable condition, and states that the results are not diagnostic.

    Achieved whenLabcorp or an outside group publishes how many buyers received a reportable finding, how many had it confirmed in clinical care, and what they did next.

    clpmag.com

    Claimed by

    Labcorp, July 2026

Open

  1. 01

    Planted-variant test files

    A public kit of made-up raw DNA files in the column layout consumer companies use (marker ID, chromosome, position, genotype), each with an answer key. Each file carries planted disease-linked variants taken from ClinVar, NCBI's public archive of variant classifications, plus common variants that ClinVar classes as benign. Run the files through open raw-data tools and AI assistants that read such files, and score each one on whether it found the planted variants, whether it called benign variants a risk, and whether it told the reader that a rare finding from a chip needs a clinical confirmation test. Use tools that run locally or models given the file directly; upload nothing to services that match DNA between users.

    The brief
    Credit · unclaimed
  2. 02

    One list of cleared tests

    A public, dated table of every genetic test the FDA has authorized for sale to consumers. Build it from openFDA's device records under the four consumer product codes (carrier screening, genetic health risk, cancer predisposition and pharmacogenetic, which predicts drug response), with each decision summary linked and its variants, limits, validation populations and current sale status filled in. Add a second list of the reports sellers offer under the rule that lets later genetic health risk reports go on sale without their own FDA review, gathered from sellers' websites.

    The brief
    Credit · unclaimed

Open 01 · the brief

Planted-variant test files

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public kit of made-up raw DNA files in the column layout consumer companies use (marker ID, chromosome, position, genotype), each with an answer key. Each file carries planted disease-linked variants taken from ClinVar, NCBI's public archive of variant classifications, plus common variants that ClinVar classes as benign. Run the files through open raw-data tools and AI assistants that read such files, and score each one on whether it found the planted variants, whether it called benign variants a risk, and whether it told the reader that a rare finding from a chip needs a clinical confirmation test. Use tools that run locally or models given the file directly; upload nothing to services that match DNA between users.

Why it is still open

The two studies above used real people's data. Ambry also found variants that a third-party interpretation service called "increased risk" and that clinical labs class as benign, but neither study set known answers for the tools that now read raw files. GitHub holds many such tools, some built to hand the file to an AI model, such as snp-dna-extract, which formats AncestryDNA raw data into an "LLM-ready report", and a GitHub search found no published test of such tools against known answers. The only synthetic raw-file generator found, Raw-dna-generator (January 2026), writes random genotypes with no planted variants and no answer key.

github.comgithub.compmc.ncbi.nlm.nih.gov

Built with what exists

ClinVar's monthly variant summary file and gnomAD's aggregate allele frequencies (counts across many people, not any one person's data) to give background genotypes realistic frequencies; Python to write the files; open tools from GitHub; AI models such as Claude and the open-weight Mistral Large 4 (see the Note "A trillion open parameters"), each run with the same fixed prompt.

  1. First step

    Write 10 files, each with three ClinVar pathogenic and three ClinVar benign common variants planted among ordinary markers, and run them through two open tools this month.

  2. Done looks like

    100 files and their answer keys published under an open license; at least 10 tools or models scored on three published rates (planted variants found, benign variants called a risk, rare findings sent with a confirmation warning); the same files rerun three months later on each tool's newest version.

Open 02 · the brief

One list of cleared tests

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public, dated table of every genetic test the FDA has authorized for sale to consumers. Build it from openFDA's device records under the four consumer product codes (carrier screening, genetic health risk, cancer predisposition and pharmacogenetic, which predicts drug response), with each decision summary linked and its variants, limits, validation populations and current sale status filled in. Add a second list of the reports sellers offer under the rule that lets later genetic health risk reports go on sale without their own FDA review, gathered from sellers' websites.

Why it is still open

The FDA's consumer-tests page lists four 23andMe products and was last updated on 20 December 2019. openFDA holds 13 records under those four codes through January 2025, among them Ancestry's Factor V Leiden test (August 2020), a Helix Alzheimer's app (December 2020), Grifols' AlphaID At Home (October 2022) and Targeted Genomics' GlutenID (January 2025). No public list shows the reports sold under the exemption.

fda.govapi.fda.gov

Built with what exists

openFDA's free device API; FDA decision summaries on accessdata.fda.gov, turned into text with pdftotext; an AI model such as Claude to fill the table's fields from each summary, with every field checked by a person against the page it came from; dated snapshots of sellers' report pages.

  1. First step

    Pull the four product codes from openFDA and enter the first five decision summaries in the table.

  2. Done looks like

    every authorization under the four codes in the table, each with its source link; a second reader agrees with at least 95% of fields in a random 20% sample; at least 10 sellers' exempt reports listed; the table refreshed every quarter for a year, with changes logged.