MIT researchers unveiled an AI model, IRIS, that identifies distinctive gene-expression “fingerprints” left by cellular signaling pathways, enabling scientists to reconstruct how cells communicate and decide their fates during development. Trained via perturbation screens and leveraging transfer learning, IRIS generalized across cell types and accurately mapped pathway activity in mouse embryos, including signals guiding lung development—predictions later confirmed experimentally. The Nature Methods study promises faster, more scalable design of organoids and stem-cell–based therapies, potentially improving disease models and drug testing. The approach could streamline regenerative medicine by replacing laborious pathway-by-pathway experiments with computational inference at single-cell resolution.
Related articles:
– CellChat: Inference of cell–cell communication from single-cell data
– RNA velocity of single cells
– NicheNet: modeling intercellular communication by linking ligands to target genes
– CellPhoneDB: inferring cell–cell communication from single-cell transcriptomes
– A single-cell molecular map of mouse gastrulation and early organogenesis
