Industries / Health / Healthcare / Medical labs
Genetic testing and diagnostics
Genetic testing labs and test makers read DNA and other inherited markers, from a newborn's heel-prick blood, a patient's blood or a tumor, to diagnose disease, screen babies and parents, and choose treatments.
1976
Y. W. Kan, M. S. Golbus and A. M. Dozy at the University of California, San Francisco, counted the alpha-globin genes in DNA from cells grown out of amniotic fluid and diagnosed alpha-thalassemia, an inherited blood disorder, in a pregnancy at risk; cord blood confirmed it at birth. The Lasker Foundation dates this first prenatal diagnosis by DNA to 1976, and UCSF calls Kan the first to diagnose a human disease using DNA.
42
Milestones achieved across 6 domains.
10
To watch for: announced, not yet confirmed outside.
12
Open deliverables. Nobody has shipped them.
In one minute
Where it started
Robert Guthrie's 1963 blood-spot test for phenylketonuria let state labs look for an inherited disease in every newborn, and in 1976 Y. W. Kan's team at UCSF diagnosed alpha-thalassemia in a pregnancy from the DNA itself. Carrier screening for Tay-Sachs disease, offered to young adults from 1970, showed that testing healthy people needs counseling as much as chemistry. In 2009 a Yale team made what its authors believed was the first diagnosis from an exome, the 1% of DNA that codes for proteins.
Where it stands
Exomes and genomes are now routine first tests for sick children: Baylor's lab found a cause in 25% of its first 250 exomes, and a genome has been read in under four hours. The hard part is deciding what a change means: nine labs given the same 99 variants agreed only 34% of the time, and in one lab's data 91% of uncertain results that were later reclassified turned out harmless. Consumer tests have reached tens of millions of people without a doctor, but a chip that checks fixed positions confirmed only 4.2% of its positive BRCA calls in one large study, and a 2023 breach at 23andMe exposed data on many users whose own accounts were never entered.
What is open
Most open jobs are public tables that nobody has built: which labs handle genes with near-identical copies, how much of ClinVar each exome kit covers, why labs disagree on each conflicting variant, and what every registered test says its result means. Others are made-up files with known answers, for secondary-findings software, consumer raw-data apps and newborn screening classifiers, and dated archives of DNA companies' privacy terms and police-request figures. None needs a real person's DNA; each uses public registries, variant-level databases, regulators' records or simulated data.
Domains
AchievedWatch forOpen
Domain
Progress
Last achieved
Watch for
Open
Reading a patient's DNA
Turning a blood sample into a readout of a patient's genes, exome (the roughly 1% of DNA that codes for proteins) or whole genome, accurately enough and fast enough for a doctor to act on.
7 · 2 · 2
Last achievedWider found more, narrower faster2023
Watch forOne long-read test for all
2
openDeciding what a change means
Judging whether a DNA change found in a patient causes disease, does not, or cannot yet be called, and keeping that judgment current as evidence comes in.
7 · 2 · 2
Last achievedUncertainty falls unevenly2024
Watch forAI scores for every letter
2
openBuilding and approving tests
Designing a genetic or other marker test, showing that it measures what it claims and that its result means something for patients, and getting it approved to sell or use.
7 · 1 · 2
Last achievedRules for inherited cancer panels2023
Watch forGene-guided prescribing by family doctors
2
openNewborn screening
Testing a few drops of blood from every baby, in the first days of life, for rare conditions that can be treated if they are found before symptoms start.
7 · 2 · 2
Last achievedThe panel's review body ends2025
Watch forGenomes for 100,000 babies
2
openTests 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 · 1 · 2
Last achievedChips miss rare variants2021
Watch forSequencing without a referral
2
openPrivacy
Who may see, keep, sell or search genetic data, and what laws, courts and regulators have said about it; a person's DNA also says something about their relatives.
7 · 2 · 2
Last achievedAustralia bars insurers' use2026
Watch forA pledge to keep policies
2
open
Advances to build with
3 research notes

Claude Fable 5.2
Three working devices
Anthropic says its AI model Claude Fable 5.2 designed three prototype devices that worked without electrical modifications, with no outside test reported.

Mistral Large 4
A trillion open parameters
Mistral published its Large 4 AI model, a trillion numbers by its count, free to download and sell, though firms that rent out models need a separate license.

Muse Ember
The jobs agents repeat
Meta's Muse Ember matched Claude Haiku 5.5 on an outside firm's tests at a third of the cost per task, though both list the same prices for text in and out.
Open in genetic testing and diagnostics
3 of 12 open deliverables
Open · Reading a patient's DNA
Look-alike genes, who reads them
A public table of the disease genes that short-read sequencing reads poorly because a near-identical copy sits elsewhere in the genome, such as PMS2, SMN1, GBA1 and CYP21A2, matched against every clinical test in the NIH Genetic Testing Registry (GTR) that lists them. For each test and gene it records whether the lab names a method that separates the copies (long-range PCR, MLPA copy-number testing, long reads or a dedicated caller), says the gene is only partly covered, or says nothing.
Built with what exists
GTR's public files: gtr_ftp.xml.gz, which holds each test's methods and was updated on 18 November 2026, and test_condition_gene.txt, which lists each test's genes; the Mandelker and Genome in a Bottle gene lists; a small extraction model such as Meta's Muse Ember to read free-text method descriptions (see the Note "The jobs agents repeat"), with every label checked against the text; Python and DuckDB. No patient data is involved.
Open · Deciding what a change means
A secondary findings answer key
A public set of made-up VCF files, the standard variant lists a sequencing pipeline produces, drawn from simulated genomes of no real person, with an answer key giving the correct secondary-findings report for each file under the ACMG list (version 3.3). The files are built to test the list's gene-by-gene rules, and the open tools that automate the search are scored against them.
Built with what exists
The ACMG SF v3.3 policy statement and the public table of it; ClinVar; gnomAD allele frequencies to draw background genotypes at random, as one open project already does with older ExAC frequencies; bcftools; the open tools SFtool from Babelomics and ACMG_SF_Classifier; Claude Fable 5.2 or another coding agent to write the generator and its tests (see the Note "Three working devices").
Open · Building and approving tests
Clinical validity, as registered
A public table, rebuilt each month from the full data file of the NIH Genetic Testing Registry (GTR), of what every registered clinical test says about its clinical validity, meaning how well a result predicts the condition. For each test it records whether the section is filled in, whether it gives a number (detection rate, sensitivity, positive predictive value), whether it cites a study, and whether the number matches the cited study. Results are grouped by lab and by test type: inherited cancer, carrier, prenatal, pharmacogenomic and others.
Built with what exists
GTR's full XML file (gtr_ftp.xml.gz, 176 MB) and its monthly archive; Python with lxml; a current AI model, such as Claude or Mistral Large 4 with open weights (see the Note "A trillion open parameters"), to sort each free-text statement into a fixed schema and pull out its number and citation; the Europe PMC API to fetch the cited abstracts for checking.
Where it sits
Broader