Most hospital tests for infection answer a narrow question: is this particular bacterium or virus present? Metagenomic testing asks a wider one: what is in this sample at all? It reads all the genetic material, human and microbial, and matches it against databases of known organisms. The method helped identify the new coronavirus in Wuhan in late 2019, and it is now moving into routine hospital care. Published results show both what it can do and why it is not yet a standard test.
The clinical side of the story was told this month by BBC Future, which interviewed researchers in the U.S. and Singapore. Here we look at the published numbers behind it.
Why doctors want it
Some patients arrive with symptoms that fit many causes. Inflammation of the brain or its lining, for example, can come from a virus, a bacterium, a fungus, a parasite or the immune system attacking the body, and the treatments differ. Standard tests check for a list of usual suspects one at a time. If the cause is not on the list, it can be missed.
A metagenomic test does not need a list. Charles Chiu’s team at the University of California, San Francisco, has run such a test on cerebrospinal fluid, the liquid around the brain and spine, since 2016. In a study published in Nature Medicine in November 2024, the team reported on nearly 5,000 samples tested through 2023. The test correctly identified the cause in 86% of neurological infections, and Chiu said it “performs better than any other category of test for neurologic infections.” One case he described to BBC Future involved a woman with meningitis and liver failure who was due for a liver transplant; the test found hepatitis E, a treatable virus that U.S. hospitals do not routinely look for, and she recovered on antiviral drugs without the transplant.
The test has received the U.S. Food and Drug Administration’s breakthrough device designation, which speeds up review but is not an approval. It is offered commercially by Delve Bio, a company the researchers co-founded.
What a London hospital learned
The most detailed public data on everyday use comes from Guy’s and St Thomas’ NHS Foundation Trust in London. Its team ran a metagenomics service for patients on ventilators with suspected chest infections across three intensive care units and published the results in the American Journal of Respiratory and Critical Care Medicine.
- Speed: the median time from sample to result was 6.7 hours, and most results came back the same day. The standard tests it was compared with take about three days.
- Accuracy: compared with culture, the test caught 93% of relevant pathogens.
- Impact: 48% of results changed antibiotic treatment, about half by adding or strengthening drugs and half by scaling them back.
- Surprises: the test found Legionella bacteria picked up in the ICU itself. The same strain was traced to a hand-basin tap, and the hospital fitted new water filters.
The paper is also frank about the limits. The team could process only three samples a day because of staff availability. About 12% of samples failed quality checks. The workflow did not detect RNA viruses, which include flu and coronaviruses. It was a single-hospital pilot, and the authors wrote that “multicenter studies are now required.”
After the trial, NHS England and the Department for Science, Innovation and Technology put about £3 million into extending the service to ten hospitals in England over two years.
Where AI comes in
Sequencing machines have become fast. The harder part is interpreting what they read. A sample contains mostly human DNA, plus harmless microbes many people carry and traces of contamination from the lab. A patient may have several infections, only one of which matters. Researchers told BBC Future they are using AI tools to filter out that noise and to read how the patient’s own genes are behaving, which can show whether the immune system is actually fighting an organism or simply living with it.
In Singapore, Niranjan Nagarajan of the National University of Singapore is testing whether AI-assisted screening can speed up treatment for people with weak immune systems, such as transplant and cancer patients, who are more likely to die from sepsis. Labs there also run the tests on sewage and air samples to look for new viruses, an idea that gained support after traces of Covid-19 were found in Italian wastewater from before the country’s first confirmed cases.
Using AI to read large biological datasets is becoming common; we covered another example in an AI finding that could lead to new treatments for genetic diseases.
Our take
The science of metagenomic testing is no longer the main obstacle; logistics are. A test that changes treatment for half of ICU patients but can handle three samples a day and loses one in eight to quality problems is a research service, not yet a hospital routine. The useful next numbers will come from the ten-hospital NHS rollout: how many samples a day each site can run, how often results arrive in time to matter, and what each test costs. If those numbers hold up, the case for making this a standard test for the sickest patients becomes hard to argue with.