During embryogenesis, a cluster of stem cells differentiates into specialized types such as brain, liver, or muscle cells and self‑organizes into three‑dimensional tissues. A cell’s fate is dictated by which genes are switched on or off, and these gene‑activity patterns are constantly reshaped by chemical signals exchanged with neighboring cells. The signals travel through multi‑step signaling pathways that translate external cues into specific transcriptional changes. Historically, researchers assumed that the effect of a given pathway varied dramatically across cell types, forcing them to map each pathway separately—a labor‑intensive endeavor.
A team led by MIT’s Pulin Li introduced a machine‑learning model called IRIS that revealed each signaling pathway leaves a unique gene‑expression “fingerprint.” Crucially, the same pathway produces highly consistent fingerprints across diverse cell types. IRIS scans single‑cell transcriptomes for these fingerprints, inferring which signals a cell received at various developmental stages, even for cell types it has never seen. The core of the model is a neural‑network‑based pattern recognizer trained on thousands of human embryonic stem cells exposed to dozens of combinations of six major pathways, creating a comprehensive signaling‑behavior atlas. Validation on mouse embryos at gastrulation showed IRIS could accurately predict the spatial‑temporal activation of pathways in progenitors of heart, gut, muscle, and spinal cord.
The model also enables reverse engineering of signal combinations that drive specific cell fates—for example, the signals required for a lung‑development‑critical cell type. Mouse experiments confirmed IRIS’s predictions, paving the way for more reliable generation of lung organoids for disease modeling and drug testing. IRIS exemplifies transfer learning in biology: knowledge acquired from one cell context transfers to others, dramatically speeding up the reconstruction of cellular communication histories. Looking ahead, this approach could accelerate stem‑cell engineering, organoid construction, and regenerative‑medicine research.
Review: By turning complex signaling cascades into reusable gene‑expression fingerprints, IRIS offers a scalable shortcut to map developmental communication and to program stem cells with unprecedented precision.