Industries / Health / Healthcare / Medical labs
Genetic testing
Reading a person's chromosomes and DNA, or a tumor's, to find changes that cause disease, predict it or guide treatment.
7
Milestones achieved, from 1959 to 2022.
3
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1959
An extra chromosome 21
In 1956 J. H. Tjio and A. Levan in Lund, Sweden, counted 46 human chromosomes, ending more than 30 years in which the number was thought to be 48; a historian calls this the foundation of modern human cytogenetics, the study of chromosomes. In 1959, in Raymond Turpin's lab in Paris, Marthe Gautier and Jérôme Lejeune found an extra chromosome 21 in children with Down syndrome, tying a medical condition to a chromosome change.
LessonAn accepted number can steer what people see; for 30 years observers counted 48.
Credit
J. H. Tjio and A. Levan; Marthe Gautier, Jérôme Lejeune and Raymond Turpin, with the shares of Gautier and Lejeune still debated
2013
Natural genes cannot be patented
In Association for Molecular Pathology v. Myriad Genetics, decided on 13 June 2013, the US Supreme Court held that a naturally occurring DNA segment is not patent eligible merely because it has been isolated, while cDNA, a lab-made copy, is. If valid, Myriad's patents would have given it the exclusive right to isolate a person's BRCA1 and BRCA2 genes, genes whose mutations can greatly raise the risk of breast and ovarian cancer. Isolating them is a step every test of them needs.
Credit
US Supreme Court; the Association for Molecular Pathology and the other plaintiffs
2015
Five words for a variant
The American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) set five standard terms for a DNA change: pathogenic, likely pathogenic, uncertain significance, likely benign and benign. They also set criteria for each, built from kinds of evidence such as how common the change is in healthy people, computer predictions, lab experiments and how it passes through families.
Credit
S. Richards, H. L. Rehm and the ACMG, AMP and College of American Pathologists workgroup
2016
Misread in Black patients
Using public exome data, A. K. Manrai, I. S. Kohane and colleagues found that DNA changes long reported as causes of hypertrophic cardiomyopathy, a thickened heart muscle, were common in healthy people. Several patients, all of African or unspecified ancestry, had received positive reports for changes that were later reclassified as benign. The changes were significantly more common in Black Americans than in white Americans.
LessonReference data must include the populations being tested, or their ordinary variants get called disease.
Credit
A. K. Manrai, I. S. Kohane and colleagues
2017
One tumor test, 324 genes
On 30 November 2017 the FDA approved FoundationOne CDx, a sequencing test on tumor tissue that looks for small changes and copy number changes in 324 genes, some gene rearrangements, and two whole-tumor measures (microsatellite instability and tumor mutational burden). One test serves as the approved companion test for targeted drugs in lung, skin, breast, colorectal and ovarian cancer.
Credit
Foundation Medicine; US FDA
2021
Genomes in a health service
The pilot of the UK's 100,000 Genomes Project read the whole genomes of 4,660 people from 2,183 families with 161 rare diseases, alongside care in the National Health Service. It made a genetic diagnosis in 25% of patients, and 25% of those diagnoses changed care for the patient or relatives at once. 14% of diagnoses needed a mix of research and automated analysis, often for changes that exome tests cover poorly.
Credit
D. Smedley and the 100,000 Genomes Project Pilot Investigators; Genomics England
2022
Screening is not diagnosis
Prenatal screening that reads fetal DNA fragments in the mother's blood (NIPS) has been sold in the US since 2011. In April 2022 the FDA warned that these tests can give false results, that none had been authorized, cleared or approved by the FDA, and that they are screening tests, not diagnostic ones. It said it had reports of pregnancies ended on screening results alone.
LessonA screening result needs a confirming test and a plain explanation before anyone acts on it.
Credit
US FDA
Watch for
A sharper splicing predictor
Splicing is the cell's cutting and joining of a gene's message, and DNA changes that disturb it are easy to miss. Illumina says its new AI model, SpliceAI2, found 17% more disease-relevant splice variants than other splicing models in Genomics England data, by its own preprint. The model is for research use, and its weights are free only for non-commercial use.
Achieved whena group outside Illumina reproduces the gain on patients Illumina did not use.
Claimed by
Illumina, October 2026
AI ranks the causal variant
Pure Genetic Medical and Breakthrough Genomics say their Virtual Geneticist software put the diagnostic DNA change in its top 10 candidates in 99% of diagnosed cases, in about 800 exomes at BC Children's Hospital in Canada. They say re-analysis raised the diagnosis rate from about 50% to 60%, and that it ranked first of six platforms in a head-to-head test run by Baylor Genetics, for which no figures are given.
Achieved whenBC Children's Hospital or Baylor Genetics publishes the evaluations with methods and numbers, or an outside group tests the software on a public set of cases.
Claimed by
Pure Genetic Medical and Breakthrough Genomics, September 2026
Agents grade gene-disease links
Before a gene goes on a test, expert panels at ClinGen, the US Clinical Genome Resource, judge how strong the evidence is that it causes the disease. Chaeeun Lee and colleagues trained a group of AI agents, led by a small open model, Qwen3-4B, to make this call on ClinGen's data. It matched the experts' final grade 75.0% of the time, against 19.5% before training, but its reasoning matched theirs much less often (0.520 on a 0 to 1 scale).
Achieved whena ClinGen expert panel or an outside group tests it on new curations made after its training.
Claimed by
Chaeeun Lee, T. Michael Yates, Pasquale Minervini and T. Ian Simpson, February 2026
Open
01
Gene panels checked against experts
An audit of every clinical test in the NIH Genetic Testing Registry (GTR) that lists genes. It flags each gene that ClinGen has classified as disputed, refuted or with no known disease link for the condition the test names, and each gene ClinGen has not yet judged. It reruns monthly and lists what changed.
The briefCredit · unclaimed02
Test genomes for rare variants
A public library of in silico reference data: sequencing reads from openly shared reference genomes (Genome in a Bottle) with clinically important DNA changes written into them by software. It covers the curated disease-causing changes in widely tested genes that have no physical reference sample. Add a one-command tool that a lab without bioinformatics staff can use to make more.
The briefCredit · unclaimed
Open 01 · the brief
Gene panels checked against experts
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An audit of every clinical test in the NIH Genetic Testing Registry (GTR) that lists genes. It flags each gene that ClinGen has classified as disputed, refuted or with no known disease link for the condition the test names, and each gene ClinGen has not yet judged. It reruns monthly and lists what changed.
Why it is still open
ClinGen's expert panels check the genes on testing panels one disease at a time. The intellectual disability and autism panel found 22 of 156 genes (14%) with only limited or disputed evidence, and said such genes are not recommended for clinical testing. ClinGen's download on 18 November 2026 held 3,735 gene-disease judgments, 201 of them disputed and 49 refuted. Literature and GitHub searches found no audit across the whole registry.
Built with what exists
GTR's public files, including a table of each test's conditions and genes, updated on 15 November 2026; GTR counts 342 registered labs, 269 of them with clinical tests; ClinGen's gene-validity download; the Mondo disease ontology to match disease names; an AI model to propose matches between GTR's condition names and ClinGen's disease entries, with each match checked against Mondo's cross-references.
First step
Join the two files for intellectual disability and autism tests, where ClinGen's panel has published, and count flagged genes per test.
Done looks like
every GTR clinical test scored, with a public table. Two clinical geneticists or genetic counselors review 100 random flags and agree with at least 95. Each lab sees its results before publication and its replies are posted. The audit reruns monthly for six months with a change list.
Open 02 · the brief
Test genomes for rare variants
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public library of in silico reference data: sequencing reads from openly shared reference genomes (Genome in a Bottle) with clinically important DNA changes written into them by software. It covers the curated disease-causing changes in widely tested genes that have no physical reference sample. Add a one-command tool that a lab without bioinformatics staff can use to make more.
Why it is still open
S. Roy, B. Funke and colleagues found public reference materials for only 29.4% of expert-curated, clinically relevant changes in widely tested genes. In their blinded pilot of 49 such changes written into data, one of three experienced labs found all of them and two missed one each. They call for central collections and say turnkey tools for labs without bioinformatics expertise are still needed. The College of American Pathologists' in silico surveys, run since 2016, cover set cancer panels.
Built with what exists
Genome in a Bottle reads for its public reference genomes; BAMSurgeon, an open tool that writes changes into sequencing files; ClinGen expert-panel variant lists and ClinVar; AI coding agents to write and test the one-command tool and its instructions.
First step
Pick one widely tested gene, list its curated disease-causing changes that lack a physical reference sample, and write five of them into one Genome in a Bottle exome.
Done looks like
files for at least 500 curated changes in 50 genes, published with a DOI. Three outside labs run their own pipelines blind and report what they detect, change by change. A person with no bioinformatics training makes one new file from the instructions in under a day.