Predictive virtual embryo systems that integrate single-cell and spatial data with artificial intelligence (AI) techniques offer a promising avenue for modeling mammalian embryogenesis across scales and could advance our fundamental understanding of development and congenital disease.

We define the virtual embryo as a fully data-backed digital twin of embryogenesis, capable of capturing multiscale dynamics from the molecular regulation of single cells to tissue patterning and organ formation. At its core, the virtual embryo bridges high-dimensional, spatiotemporal single-cell atlases of mammalian development with advanced machine learning techniques, including graph neural network models and foundation models. This framework enables the simulation and prediction of embryogenesis under normal or genetically and environmentally perturbed conditions, thus allowing the identification of molecular mechanism for developmental disorders and the mitigation of them by compensating for these perturbations.