Cells developing into tissues and organs rely on chemical messages from their neighbors. A new machine learning model called IRIS offers researchers a way to infer those past signals from patterns of gene activity, including in cell types the model has not seen before.
Gene activity records a cell’s signals
Early in development, stem cells take on specialized roles such as brain, liver, or muscle cells. Which genes are switched on or off helps determine each cell’s structure and function, but those decisions are shaped by communication with nearby cells.
Signals move through multi-step sequences called signaling pathways. These pathways translate outside messages into changes in gene activity. Reconstructing the signals a cell received could help explain how tissues develop and how those processes go wrong in disease.
Researchers have faced a difficult mapping task: if a pathway’s effects vary across cell types, then each pathway must be studied separately in every type. The new study, led by MIT assistant professor of biology and Whitehead Institute for Biomedical Research member Pulin Li and graduate student Nicholas Hutchins, reports a way around that burden.
IRIS looks for pathway fingerprints
The team found that each signaling pathway leaves a distinctive pattern of gene activity, or fingerprint. For the same pathway, that pattern remains consistent across different cell types. This gives researchers a shared signal to look for instead of requiring a separate map for every type of cell.
IRIS examines a cell’s overall gene activity and estimates which signaling pathways were likely active at particular stages of development. It is a neural network-based model trained to recognize patterns in complex data.
Li and Hutchins trained the model using experimental measurements from thousands of human embryonic stem cells. The data covered dozens of combinations of six major signaling pathways at multiple stages of development, creating an atlas of how those combinations influence cell behavior.
Li described the approach as transfer learning, comparing it to voice recognition systems trained mainly in English that can use that training to recognize other languages. In the same way, IRIS can apply pathway fingerprints learned in one set of cells to other cell types.
Testing predictions in developing embryos
The researchers tested IRIS on single cells from mouse embryos during gastrulation, when cells are rapidly branching into different fates. The model predicted when and where particular pathways would activate in cells destined to contribute to heart, gut, muscle, and spinal cord tissue.
The team also used IRIS to identify signals needed to create a cell type important for lung development. The model predicted that activating a specific signaling pathway would encourage lung-specific development, and experiments in mouse embryos confirmed that prediction.
These results suggest that IRIS can help researchers move from observing a cell’s gene activity to forming a picture of the developmental instructions it received. That could make it possible to study signaling histories across many cell types without experimentally testing every pathway in every possible type.
Potential uses in stem cells and disease research
Understanding the signals that guide a stem cell toward a particular identity could give researchers a practical roadmap for producing specific cell types in the lab. The findings may support stem cell engineering for regenerative medicine and improve organoids, which are miniature 3D models that mimic real organs.
More accurate lung tissue models could help researchers investigate asthma, lung cancer, and pulmonary fibrosis, a disease in which lung tissue becomes scarred, often without a known cause. Such models could also be used to test potential therapies and inform efforts to design regenerative treatments for damaged tissue.
The study’s open-access findings were published in Nature Methods on Sept. 8. The researchers’ approach could speed up the work of decoding cell-to-cell communication, while giving scientists a way to explore how signals shape development and disease.