A digital twin of pancreatic islet differentiation predicts cell fate

In vitro differentiation of human pluripotent stem cells into pancreatic islet organoids (SC-islets) is critical for widespread diabetes therapy but remains limited by incomplete control of cell fate specification. Despite significant insights gained from single-cell sequencing technologies, the field lacks a predictive framework to resolve cell fate specification and infer the regulatory determinants of lineage trajectories. Here, we report an integrated digital twin of pancreatic islet differentiation constructed from temporal multi-omic data spanning in vitro development. This predictive engine resolves lineage trajectories and defines cell-state-specific regulatory networks across differentiation. Analysis of over 1,000 in silico transcription factor perturbations identifies the key candidate regulators of key developmental branch points, including previously unrecognized drivers of lineage specification. Prospective perturbation experiments validate model-predicted regulators of lineage specification, establishing causal inference of cell fate decisions. We describe a predictive digital twin that enables causal inference of cell fate decisions and provides a foundation for improving SC-islet biomanufacturing and therapeutic development.