Gateway analysis reveals transient molecular programs at cell-fate transitions

Single-cell atlases have transformed our view of cell identity, but they still struggle to resolve the transient states that accompany cell-fate change. These rare interface cells are outnumbered by the stable populations on either side. As a result, the molecular programs that accompany lineage transitions are hard to detect. We developed gateway analysis, a framework that identifies cells at fate boundaries and pinpoints bell and valley genes that peak or dip there. It defines cell neighborhoods from binary mutual information (BMI) in gene on/off patterns and preserves that structure in a regularized latent model. Across four single-cell atlases spanning reprogramming, gastrulation, pancreatic endocrinogenesis, and kidney injury (3,696–126,578 cells), gateway analysis recovered rare boundary populations and the transient gene programs that distinguish them from their flanking states. These signals were missed by standard comparisons of stable endpoint states. They included epithelial remodeling at the late MET-to-iPSC interface, a gate-versus-basin partition of gastrulation regulators, a transient Cck-enriched peak at endocrine hub entry together with a candidate BH4-associated signal at a shared endocrine hilltop, and a proteostasis program at the kidney injury boundary. Orthogonal support from optimal-transport fate probabilities, known markers, and published perturbation phenotypes indicate that gateway cells mark bona fide biological transition intervals. Gateway analysis therefore provides a practical framework for detecting rare transition-state cells and the bell/valley genes that define them in single-cell atlases.