NederlandsAI copies the immune system of pregnant women and finds a candidate drug against preterm birth
Stanford researchers used digital twins of the immune system to test nine existing drugs. In the simulation, aspirin plus a stomach acid reducer delayed labour far longer than aspirin alone. A large study in pregnant women must show whether this works in real life.
One in ten children in the United States is born too early, before the 37th week of pregnancy. Worldwide, according to the World Health Organization, this affects more than 15 million babies a year. Babies born too early more often have problems with breathing and development. Even so, there are hardly any drugs that help prevent an early birth.
Researchers at Stanford Medicine have now used AI to find a new candidate: a combination of aspirin and lansoprazole, a stomach acid reducer. In computer simulations, this combination delayed labour about fifteen times longer than aspirin alone. Their study was published on 21 October 2026 in the journal Science Advances.
Why is this so difficult?
Testing new drugs in pregnant women is hard. Nobody wants to put mother and child at risk, so such studies are rare, expensive and slow. As a result, research into preterm birth lags behind.
What is known is that the mother's immune system plays a big part. Towards the end of pregnancy, inflammation signals in the body increase, and these can start labour. If the signals come too early, labour can also start too early. A drug that slows these signals at the right moment could therefore extend the pregnancy.
A digital twin of the immune system
The team, led by Brice Gaudilliere with Jakob Einhaus as first author, therefore built a computer model called SIMCO. It creates a digital twin of a pregnant woman's immune system: a copy in the computer, down to the level of individual immune cells. In it, you can try a drug without a real patient having to swallow it.
To train the model, the scientists studied blood from 19 people, some pregnant and some not. In the lab, they exposed this blood to nine drugs chosen in advance, as well as other substances and combinations, in a total of 80 different experiments. They measured how 218 million immune cells reacted. From these measurements, the AI model learned how a drug changes a cell. A second model linked these changes to the question of how long a pregnancy lasts.
The researchers then applied the model to blood from 63 other pregnant women, five of whom gave birth too early. The model's predictions matched fairly well with when these women actually gave birth. This control group is small, as the researchers say themselves.
What the model found
Of the options tested, the combination of aspirin and lansoprazole came out best. In the simulation, 87 per cent of the digital twins responded to it. In them, labour was delayed by more than 17 days on average, compared with about 1 day with aspirin alone. According to the researchers, the two drugs strengthen each other: together they dampen the immune signals that start labour. Tetrahydrofolate, a form of vitamin B9, and the drug maprotiline also came out as candidates.
Both drugs have existed for a long time. Aspirin has been studied before as a way to prevent preterm birth, and pregnant women are sometimes prescribed lansoprazole for heartburn. This makes the step to research in people smaller than with a completely new drug. In an animal model with inflammation during pregnancy, labour did indeed come later with the combination.
Next step: a study in pregnant women
The US medicines regulator, the FDA, gave permission for a study in people. In it, Stanford will test the combination in about 670 pregnant women at higher risk of an early birth. They will take a low dose of aspirin and lansoprazole every day, from week 12 to 16 of pregnancy until labour. Another group will receive the usual care. Only a study like this can show whether the effect seen in the computer also happens in real life.
This matters, because the factor of fifteen comes only from simulations. How well the combination really works, which dose is needed and whether it is safe enough is for the study to show. Pregnant women should therefore not combine these drugs themselves, but discuss such questions with their midwife or doctor.
There is a great need for new options. In the United States, one drug was specially approved to prevent repeat preterm birth, Makena. The FDA withdrew that approval after a large study showed no effect.
What does this mean for the road to the singularity?
This research shows how AI can speed up science where trials in people are difficult. Instead of testing drugs one by one in pregnant women, the researchers tried many at once in digital twins. Only the best candidate goes on to a study in people.
Because each digital twin is based on one person's blood, such a model can also show for whom a drug is likely to work. Studies can then become smaller and more targeted. If the combination works in the new study, that is not only good news for many parents and babies. It would also prove that digital twins can open a faster route to better drugs.
What is the technological singularity? Read the explanation →
