That consistency is what makes the method scale. Instead of mapping cell type by cell type, scientists can use these shared fingerprints to reconstruct signaling histories across many cell types at once.

"Think of voice recognition systems like Siri, which are trained mainly in English, but then use that training to help them recognize other languages," Li said. "This is called transfer learning, and this is why IRIS can work across many different cell types."

The findings are open-access.

How was IRIS built and tested?

IRIS is a neural network — an AI system, loosely brain-inspired, that spots patterns in complex data. It reads a cell's overall gene activity and estimates which signaling pathways were likely active at specific points in development, even in cell types it never saw during training.

The training data came from a large experiment measuring how thousands of human embryonic stem cells responded to dozens of combinations of six major signaling pathways at multiple developmental stages. The result: an atlas of how signal combinations shape cell behavior.

The team then tested IRIS on single cells from mouse embryos during gastrulation — the stage when cells rapidly branch into different fates. IRIS accurately predicted the timing and location of pathway activation in cells headed for heart, gut, muscle, and spinal cord tissue. That variation in signaling is what steers cells to build the right structures in the right places.

What could it do for medicine?

The practical payoff is control. Once researchers know the pattern of signals that pushes a stem cell toward a specific fate, they can replay those signals in the lab to steer the outcome.

The team already put that to the test. Asked to identify the signals needed to make a cell type critical for lung development, IRIS predicted that activating a specific pathway would encourage lung-specific development. Experiments in mouse embryos confirmed it — a validation that points toward more reliable, lab-grown lung tissue.

Better organoids — miniature 3D organ models — would let researchers study diseases such as asthma, lung cancer, and pulmonary fibrosis, a scarring of lung tissue that often has no known cause, and then test therapies on them. The long-term goal: regenerative treatments that repair damaged tissue.

"In these ways, IRIS is helping us decode the language cells use to talk to each other at a much faster rate than we could realistically achieve through experiments," Hutchins said.

How solid are the results?

Some caution is warranted. The mouse-embryo predictions cover one developmental stage, gastrulation, and the lung experiment ran in mouse embryos, not human tissue. Whether the fingerprints hold equally across all cell types and stages will take more work to confirm.

Still, the study opens a path toward mapping the signaling history of every cell in a mouse or human embryo at a scale traditional experiments couldn't realistically reach.

In short: cells keep a readable record of the messages they've received, and AI can now read it.