Industries / Health / Healthcare / Medical labs / Genetic testing and diagnostics
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
Milestones achieved, from 1986 to 2023.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1986
Four dyes, one gel lane
L. M. Smith, L. E. Hood and colleagues at the California Institute of Technology gave each of the four sequencing reactions, one per DNA letter, its own colored fluorescent dye. The four mixtures could then run together down a single gel tube, and a computer read the colors as the DNA pieces passed a detector, instead of a person reading bands from four separate lanes. The paper calls it partial automation of DNA sequence analysis.
Credit
L. M. Smith, L. E. Hood and colleagues, in Nature
2009
A diagnosis from an exome
M. Choi, R. P. Lifton and colleagues at Yale read the exome of a patient referred with suspected Bartter syndrome, a kidney disease that wastes salt. They found instead a change in SLC26A3, the gene for congenital chloride diarrhea, and clinical follow-up confirmed it. The authors wrote that, to their knowledge, exome or genome sequencing had not been used to make a genetic diagnosis before.
LessonTesting only the genes of the suspected disease would have missed it; five more patients with suspected Bartter syndrome and no changes in its known genes also carried SLC26A3 changes.
Credit
M. Choi, R. P. Lifton and colleagues, in PNAS
2012
A genome in 50 hours
Babies in intensive care get sicker in days, while genetic tests then took weeks. At Children's Mercy Hospital in Kansas City, C. J. Saunders, S. F. Kingsmore and colleagues, with Illumina, described whole-genome sequencing with automated analysis that gave a list of likely genetic causes in 50 hours, built as a prototype for neonatal intensive care units. In a 2018 study of 42 acutely ill infants at Rady Children's Hospital in San Diego, rapid genomes found a diagnosis in 43%, against 10% for standard genetic tests.
Credit
C. J. Saunders, S. F. Kingsmore and colleagues; L. Farnaes and colleagues at the Rady Children's Institute for Genomic Medicine
2013
Exomes diagnose one in four
Baylor College of Medicine's certified clinical lab reported on the first 250 patients whose doctors had ordered an exome, about 80% of them children with neurologic conditions. It found a likely cause in 62 patients (25%), and insurance coverage was similar to that for established genetic tests.
Credit
Y. Yang, C. M. Eng and colleagues at Baylor College of Medicine, in the New England Journal of Medicine
2015
A yardstick genome for labs
The US National Institute of Standards and Technology (NIST) released RM 8398, a reference material: DNA from one cell line whose genome had been read many times by several methods, so that a lab can count its own true and false calls against it. NIST said sequencing biases could produce hundreds of thousands of disagreements between results for the same genome. The FDA had used the draft version in November 2013 when it cleared one of the first commercial high-throughput sequencers.
Credit
NIST and the Genome in a Bottle Consortium, with the FDA; J. M. Zook, M. Salit and colleagues
2022
A genome in five hours
A Stanford-led team, working with Oxford Nanopore, Google and NVIDIA, spread one patient's DNA across 48 nanopore flow cells and analyzed the reads in the cloud as they came off the machine. One genome was sequenced in 5 hours 2 minutes, the first Guinness World Records title for fastest DNA sequencing, certified by NIST's Genome in a Bottle group, and the fastest diagnosis took 7 hours 18 minutes; 5 of 12 intensive care patients received a genetic diagnosis. In September 2025 Broad Clinical Labs, Roche and Boston Children's Hospital cut the record to 3 hours 58 minutes 59 seconds, though their intensive care results were research results and were not used for diagnosis.
med.stanford.edueuropepmc.orgguinnessworldrecords.compmc.ncbi.nlm.nih.gov
Credit
J. E. Gorzynski, S. D. Goenka, E. A. Ashley and colleagues; M. H. Wojcik and colleagues; Guinness World Records
2023
Wider found more, narrower faster
The GEMINI study tested 400 hospitalized infants at six US hospitals with both a genome and a targeted panel of newborn disease genes, each read by its own lab. Counting both disease-causing and uncertain variants, the genome found a result in 49% of infants and the panel in 27%, but the panel found 19 variants the genome did not report, and routine genome results took a median 6.1 days against 4.2 (3.3 against 4.0 for urgent cases). The labs' variant interpretations differed by 43%.
LessonThe test and the lab both shape the answer; the same baby's DNA can get different reports.
Credit
J. L. Maron, S. Kingsmore, J. M. Davis and the GEMINI investigators, in JAMA
Watch for
One long-read test for all
Long-read sequencing reads DNA in stretches of thousands of letters, which can span repeats and near-copies that short reads cannot place. Radboud university medical center and Maastricht UMC+ in the Netherlands ran long-read genomes on 1,000 samples, one year's mix of their routine germline testing, alongside standard care. They report 96.4% agreement, findings that improved or refined the diagnosis in 3.4% of cases, and 0.2% of cases where standard tests found a variant that long reads missed; modeled over a year's 15,150 cases, the gain in diagnoses was 2.5 percentage points. The centers say it can replace 15 other tests and recommend it as the first-choice test everywhere.
Achieved whena lab outside these two centers runs long reads as its first test on its own routine cases and publishes agreement, diagnoses, turnaround and cost.
Claimed by
Radboudumc and Maastricht UMC+ (T. J. J. de Bitter, L. E. L. M. Vissers, A. Hoischen and colleagues), June 2026
Long reads for look-alike genes
Some disease genes have near-identical copies (pseudogenes) elsewhere in the genome, and short reads cannot tell the copies apart. In 86 people with 125 already-known variants across 11 such regions, PacBio HiFi long reads plus Paraphase, a caller for these regions published by PacBio, found all 125; standard long-read callers found 95.
Achieved whena prospective clinical study, or a lab with no tie to PacBio, finds the same sensitivity on cases not chosen for having a known variant.
Claimed by
B. van der Sanden, A. Hoischen and the HiFi Solves EMEA Consortium, June 2026
Open
01
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.
The briefCredit · unclaimed02
Checking exome kits' ClinVar claims
A public table that overlays the target files published for current exome capture kits (sets of probes that pull the coding DNA out of a sample before sequencing) on every change in ClinVar classified as pathogenic or likely pathogenic. Per kit and gene, it reports the share of those changes inside the targets with 0, 10 and 25 letters of margin, split into coding, splice-region and deeper intron changes, and it reruns with each monthly ClinVar release.
The briefCredit · unclaimed
Open 01 · the brief
Look-alike genes, who reads them
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
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.
Why it is still open
In 2016 D. Mandelker, B. Funke and colleagues published a catalog of such regions to guide test design and extra assays, and in 2022 the Genome in a Bottle Consortium counted 395 challenging medically relevant genes left out of its earlier benchmarks. A June 2026 study found all 125 known variants in 11 such regions with long reads (see Watch for above). Literature, web and GitHub searches found no survey of what clinical labs say they do for these genes. GTR counts 342 registered labs, 269 of them with clinical tests.
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.
First step
Pull every GTR clinical test that lists PMS2 and label the stated methods of 20 of them by hand, to set the labeling rules.
Done looks like
Every GTR clinical test that lists at least one of 50 named look-alike genes is labeled. Two clinical molecular geneticists check 100 random labels and agree with at least 95. Each lab sees its rows before publication and its replies are posted. The table is rerun after three months with a list of changes.
Open 02 · the brief
Checking exome kits' ClinVar claims
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table that overlays the target files published for current exome capture kits (sets of probes that pull the coding DNA out of a sample before sequencing) on every change in ClinVar classified as pathogenic or likely pathogenic. Per kit and gene, it reports the share of those changes inside the targets with 0, 10 and 25 letters of margin, split into coding, splice-region and deeper intron changes, and it reruns with each monthly ClinVar release.
Why it is still open
In 2014 J. Y. Park, P. Fortina and colleagues checked three exome tests against 18,336 disease-variant positions listed in HGMD, a paid database, for the 56 genes then on ACMG's secondary-findings list, and found the clinical exome covered more than half of the positions inadequately in 7 genes; that work used one paid list, older kits and 56 genes. Kit makers publish their own figures: Twist Bioscience's 2022 poster says its expanded exome has "ClinVar coverage of over 99%". An independent 2025 comparison of off-the-shelf kits in NAR Genomics and Bioinformatics measured how well kits capture their own targets and the standard coding regions, not ClinVar's disease-causing sites. An open tool posted in May 2026, NeuropediaLAB's Exome Coverage Advisor, checks single genes against kit files a user uploads but publishes no table.
Built with what exists
The makers' published target BED files; ClinVar's monthly release files, archived back to 2024 and earlier; bedtools and Python; an AI coding agent such as Claude Fable 5.2 to write and test the scripts (see the Note "Three working devices"). No patient data is involved.
First step
Download one maker's current exome target file and the November 2026 ClinVar release, and compute the share of pathogenic and likely pathogenic sites inside the targets for the genes on ACMG's secondary-findings list.
Done looks like
A dated table for at least six current kits from at least four makers, every number reproducible from a published script. It is rerun on three monthly ClinVar releases. Each maker is invited to respond before publication. An outside bioinformatician recomputes 20 genes independently and gets the same numbers.