Cells leave a fixed fingerprint for each signaling pathway, letting one model read their signal history across cell types
Synopsis
Each signaling pathway leaves a consistent fingerprint of gene activity across cell types, so signaling history no longer has to be reconstructed separately for every cell type. IRIS is a neural-network model that reads a cell's overall gene activity and estimates which pathways were switched on at each developmental stage. Tested on single cells from mouse embryos during gastrulation, it predicted when and where pathways activate in heart, gut, muscle and spinal cord lineages, and its prediction for a lung-development cell type was confirmed in mouse embryo experiments.
Interpretation
A given signaling pathway leaves a recognizable fingerprint of gene activity that holds across cell types, so signaling history can be reconstructed across types instead of mapped one type at a time. Researchers had assumed pathway effects vary widely by cell type, which meant testing every pathway experimentally in every cell type. The consistency was established from responses of thousands of human embryonic stem cells to dozens of combinations of six major pathways at multiple developmental stages, then tested on single cells from mouse embryos during gastrulation.
IRIS infers which signaling pathways were likely active at particular developmental times by reading a cell's overall gene activity. The model is a neural network, an AI system that recognizes patterns in complex data, and it works across cell types through transfer learning. Training covered thousands of human embryonic stem cells responding to dozens of combinations of six pathways; testing used single cells from mouse embryos during gastrulation, where the model predicted pathway activation timing and location for cells destined for heart, gut, muscle and spinal cord tissue.
The model's predicted signal combinations can point to concrete interventions on cell fate. For a cell type critical to lung development, the model predicted that activating a specific signaling pathway would encourage lung-specific development. Experiments in mouse embryos confirmed that prediction, indicating the predicted signal combinations can guide generation of lung tissue models in the lab.
Perspective
The report rests on an open-access study published in Nature Methods, covering a signaling-response atlas in human embryonic stem cells and single cells from mouse embryos during gastrulation. Predicted pathway activation locations were tested in mouse embryos, and the lung-development prediction was separately confirmed in mouse embryo experiments. It suits researchers and engineers working on cell signaling, stem cell engineering and organoids.
The conclusion that pathway fingerprints are consistent across cell types was drawn within the six major pathways and the cell types tested. Inference on unseen cell types relies on transfer learning, and accurate prediction in mouse gastrulation is the validation setting reported so far.