BRENNEN

Industries / Health / Healthcare / Fertility clinics

Genetic testing

Testing DNA from a few cells of an embryo, or from the fluid it grows in, for an inherited disease or for missing and extra chromosomes, before an embryo is chosen for transfer to the uterus.

7

Milestones achieved, from 1990 to 2026.

2

To watch for: announced, not yet confirmed outside.

3

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1990

    First tests on embryo cells

    A team at Hammersmith Hospital in London took one cell from each embryo at the six- to eight-cell stage and tested it for a stretch of DNA found only on the Y chromosome. Two couples at risk of passing on X-linked diseases, which mostly affect boys, had female embryos transferred, and both women became pregnant with twins; the paper calls these the first established pregnancies from the procedure. In 1992 the same group reported the birth of a girl after embryos were tested for the deletion behind most cystic fibrosis, showing the method could work for a single-gene disease.

    europepmc.orgeuropepmc.org

    Credit

    A. H. Handyside, R. M. Winston and colleagues, Hammersmith Hospital; M. R. Hughes on the 1992 paper

  2. 2007

    Screening lowered birth rates

    Preimplantation genetic screening (PGS) counted a few chromosomes in one cell taken from a day-3 embryo, and was offered to older women in the hope of raising their chances. A double-blind trial led from the Academic Medical Center in Amsterdam randomly assigned 408 women aged 35 to 41 to IVF with or without screening. Live births fell to 24% of women with screening, against 35% without (rate ratio 0.68).

    LessonA test that removes cells and discards embryos can lower births while sounding helpful, and only a randomized trial showed it.

    europepmc.org

    Credit

    Sebastiaan Mastenbroek, Fulco van der Veen and colleagues

  3. 2018

    Edited babies, then prison

    In November 2018 He Jiankui said he had used CRISPR to change the CCR5 gene in IVF embryos, hoping to make the children resistant to HIV, and that twin girls had been born. The changes might be passed on to their own children. MIT Technology Review and the Associated Press broke the story the day before an international genome-editing summit in Hong Kong, as the Stanford law professor Henry T. Greely recounts. In December 2019 He and two embryologists were sentenced to prison and fined; He received three years.

    LessonAn ordinary IVF lab has the tools to edit embryos for birth, so the safeguard is review, licensing and law, not technical difficulty.

    pmc.ncbi.nlm.nih.goveuropepmc.orgeuropepmc.org

    Credit

    He Jiankui (the experiment); Henry T. Greely, J. Johnston and H. Jia (accounts)

  4. 2021

    Screening adds no births

    Testing moved to a few cells from the outer layer of a day-5 embryo, a blastocyst, checked for all chromosomes at once (PGT-A). A 2013 trial of 155 couples at one New Jersey practice reported delivery rates of 84.7% with testing and 67.5% without. Larger trials disagreed. The STAR trial of 661 women aged 25 to 40 found ongoing pregnancy rates per transfer of 50% and 46%, not a significant difference (2019). A multicenter trial in China of 1,212 women aged 20 to 37, each with at least three good blastocysts, found cumulative live births of 77.2% with PGT-A and 81.8% without (2021).

    LessonBetter results per transfer do not mean more babies per patient, because testing can rule out embryos that would have become babies.

    europepmc.orgeuropepmc.orgeuropepmc.org

    Credit

    Richard T. Scott and colleagues; Santiago Munné and the STAR Study Group; Junhao Yan, Zi-Jiang Chen and colleagues, Shandong University

  5. 2021

    Mosaic embryos can be healthy

    A "mosaic" result means the biopsy held a mix of normal and abnormal cells, and such embryos were usually discarded. In 2015 Ermanno Greco, Maria Giulia Minasi and Francesco Fiorentino of the European Hospital in Rome reported healthy babies after transferring mosaic blastocysts. In 2021 a double-blind non-selection trial, in which embryos were transferred without anyone knowing their test results, found the same live-birth and miscarriage rates for 484 normal, 282 low-grade mosaic and 131 medium-grade mosaic embryos, and no mosaicism in the pregnancies or newborns.

    LessonA biopsy of a few outer cells does not settle what the whole embryo is, so discarding embryos on that result loses some that would become healthy babies.

    europepmc.orgeuropepmc.org

    Credit

    Ermanno Greco, Maria Giulia Minasi and Francesco Fiorentino; Antonio Capalbo, Carlos Simón and colleagues, Igenomix

  6. 2022

    Geneticists reject embryo risk scores

    A polygenic risk score adds up thousands of small DNA differences to estimate the risk of a common disease or the likely size of a trait. A 2019 study in Cell estimated that screening embryos with the scores then available would gain on average about 2.5 cm of height or 2.5 IQ points, and that in large families most children with the top height score were not the tallest. In 2022 the European Society of Human Genetics called the use of these scores in embryo testing "an unproven, unethical practice", noting that no clinical research protocol had tested it, and in 2024 the American College of Medical Genetics and Genomics (ACMG) issued its own statement on their clinical use.

    europepmc.orgpmc.ncbi.nlm.nih.goveuropepmc.org

    Credit

    Ehud Karavani, Shai Carmi and colleagues, Hebrew University of Jerusalem; Francesca Forzano and the European Society of Human Genetics

  7. 2026

    No gain from culture medium

    Non-invasive PGT-A (niPGT-A) skips the biopsy and reads the DNA an embryo sheds into the liquid it grows in. A double-blind trial at 13 fertility centers in China randomly assigned 1,189 women aged 35 to 42. Ongoing pregnancy was 38.7% when the embryo was chosen by niPGT-A and 36.8% when it was chosen by appearance, not a significant difference, though early miscarriage was lower (18.0% against 25.2%). The authors wrote that the results "do not provide a basis for recommending routine use" in this group.

    europepmc.org

    Credit

    Jin Huang, Jie Qiao and colleagues, Peking University Third Hospital (Nature Communications, September 2026)

Watch for

  1. Disease scores checked in families

    Herasight said it had built polygenic scores for 17 diseases, including type 2 diabetes, breast cancer and Alzheimer's disease, and had checked them within families, using siblings and parents. This is the comparison that matters, because embryos from one couple are siblings. It also said that the risk reductions advertised by Genomic Prediction, Orchid and Nucleus were inflated.

    Achieved whena group with no tie to the company reproduces the within-family accuracy for at least one disease in a cohort the company did not use, in a peer-reviewed paper.

    herasight.substack.com

    Claimed by

    Herasight, July 2025

  2. Embryo trait scores on sale

    NPR reported that Orchid Health scores embryos for disease risk only, Herasight also predicts height, BMI, longevity and IQ, and Nucleus Genomics adds traits such as eye color and left-handedness, at up to $50,000 plus the cost of IVF. The companies say they have scored thousands of embryos for hundreds of prospective parents. Susan Klugman, a former ACMG president, told NPR that "current evidence doesn't support their accuracy, their safety or their clinical value".

    Achieved whena regulator or a peer-reviewed study with no company funding confirms that the scores predict differences between siblings as advertised.

    npr.org

    Claimed by

    Orchid Health, Herasight and Nucleus Genomics, as reported by NPR, May 2026

Open

  1. 01

    Named rules for mosaic calls

    A public table that names each PGT-A testing lab and the rules it states for calling a biopsy normal, mosaic or abnormal: the cut-off percentages of abnormal cells, the smallest piece of chromosome it reports, whether it reports mosaic results at all, the sequencing platform, and the date and document each rule comes from. Every value cites a public document, such as the lab's website, a sample report, a validation paper or a patient leaflet.

    The brief
    Credit · unclaimed
  2. 02

    Synthetic embryo test sets

    An open generator of made-up families: simulated parents' genomes carrying known disease variants, simulated embryos with known crossovers and known whole or mosaic chromosome errors, and simulated sequencing reads from five-cell biopsies with the uneven coverage and dropped alleles that amplifying a few cells causes. Each file comes with its true answer, so pipelines for single-gene disease (PGT-M) and chromosome counts (PGT-A) can be scored without any patient's DNA.

    The brief
    Credit · unclaimed
  3. 03

    Risk claims tested against models

    A dated public table of every number that polygenic embryo-screening companies publish about lowering disease risk, each with the company's own stated inputs (number of embryos, disease prevalence, accuracy of the score), the result an independent open model gives for the same inputs, and the gap between the two.

    The brief
    Credit · unclaimed

Open 01 · the brief

Named rules for mosaic calls

Credit · unclaimed

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

What to build

A public table that names each PGT-A testing lab and the rules it states for calling a biopsy normal, mosaic or abnormal: the cut-off percentages of abnormal cells, the smallest piece of chromosome it reports, whether it reports mosaic results at all, the sequencing platform, and the date and document each rule comes from. Every value cites a public document, such as the lab's website, a sample report, a validation paper or a patient leaflet.

Why it is still open

A 2024 study of 36,395 biopsies tested by eight labs found that the seven labs reporting mosaicism called it in 3.1% to 25.0% of biopsies, and that cycles tested by the lab that reported none were more likely to have at least one usable embryo (odds ratio 1.30). The labs were coded A to H, not named. A 2024 modeling study notes that reporting criteria for partial-chromosome changes "vary substantially across clinics". No public table names labs and their rules.

europepmc.orgpmc.ncbi.nlm.nih.gov

Built with what exists

an AI agent that browses the web and reads PDFs, such as Claude driving the open-source Browser Use library (see the Note "Deleting 80% to start over"), to find each lab's documents and quote the passage behind each value; a spreadsheet; two people checking every row against its source. The advance is agents that can read many vendor documents and cite the line each extracted value came from.

  1. First step

    List 20 PGT-A providers from lab websites and national society directories, and extract the stated rules of five of them by hand to fix the table's columns.

  2. Done looks like

    At least 20 labs in at least five countries, every value linked to a dated public document and its quoted passage; two reviewers agree on at least 95% of values before reconciling; labs that publish no rules are listed as such; the table is released under an open license and rechecked after six months.

Open 02 · the brief

Synthetic embryo test sets

Credit · unclaimed

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

What to build

An open generator of made-up families: simulated parents' genomes carrying known disease variants, simulated embryos with known crossovers and known whole or mosaic chromosome errors, and simulated sequencing reads from five-cell biopsies with the uneven coverage and dropped alleles that amplifying a few cells causes. Each file comes with its true answer, so pipelines for single-gene disease (PGT-M) and chromosome counts (PGT-A) can be scored without any patient's DNA.

Why it is still open

Physical reference samples for PGT-M exist for one disease: in 2024 China's National Institutes for Food and Drug Control and partners built a 16-sample thalassemia panel from cell lines, noting "a lack of reference materials". The Tessera R package simulates where abnormal cells sit in an embryo and which cells a biopsy takes, but stops at cells, not sequencing reads. LD-PGTA (2021) simulated low-coverage reads for trisomies from 1000 Genomes haplotypes to test its own method, assuming even coverage. No public generator joins families, embryos and reads.

europepmc.orgpmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

Built with what exists

msprime, which simulates genomes for made-up populations; the Tessera R package; a read simulator such as wgsim or ART; open copy-number callers such as CNVkit as the first tools to score; a coding agent such as GitHub Copilot (see the Note "Effort, not model") to write and test the glue code. The advance is coding agents that can assemble and test such a pipeline from existing tools in weeks.

  1. First step

    Simulate one family with msprime, plant a known disease variant, make four embryos with known crossovers, and generate reads for one biopsy each.

  2. Done looks like

    1,000 synthetic families and 10,000 embryos covering at least 20 inherited conditions and mosaic levels from 10% to 90%, released with truth tables; at least two open pipelines scored, with error rates per condition and per mosaic level; a second team regenerates the set from the code and gets identical truth files.

Open 03 · the brief

Risk claims tested against models

Credit · unclaimed

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

What to build

A dated public table of every number that polygenic embryo-screening companies publish about lowering disease risk, each with the company's own stated inputs (number of embryos, disease prevalence, accuracy of the score), the result an independent open model gives for the same inputs, and the gap between the two.

Why it is still open

PEStimate, an open app from Shai Carmi's group (Bioinformatics, June 2026), predicts a child's risk after screening and finds that risk reductions "have been previously overestimated", because the chosen embryo may not lead to a birth. A 2025 comparison on the blog Astral Codex Ten tabled the companies' reported risk reductions "taking all claims at face value". A 2024 article in the Harvard Journal of Law & Technology examined truth-in-advertising problems with these tests, and an Australian group judged in February 2026 that the predictions are "extremely uncertain". None ran each published claim through an independent model.

europepmc.orggithub.comastralcodexten.comeuropepmc.orgfindanexpert.unimelb.edu.au

Built with what exists

PEStimate (open source); an AI agent that browses company websites, white papers and ads and saves an archived copy of each claim, such as Claude with Browser Use (see the Note "Deleting 80% to start over"); a spreadsheet. The advance is agents that can collect and quote claims across many pages and keep the archive current.

  1. First step

    Collect every numeric risk-reduction claim from two companies' public pages and run the first five through PEStimate.

  2. Done looks like

    Claims from at least five companies, each row with its link, archived copy, the inputs used and PEStimate's result; two people check every row; the table is updated every three months for a year; each company is sent its rows 30 days before publication and its replies are published with them.