Quantitative comparison of in vitro and in vivo embryogenesis at a single cell resolution

Stem cell engineering is a powerful platform to address numerous challenges in regenerative medicine and developmental biology. Typically, engineered populations are derived by exposing pluripotent stem cells to a series of signaling cues meant to recapitulate developmental milestones, such as the induction of the primitive streak. Morphologic, genetic, transcriptomic, and functional differences between fully differentiated in vivo and engineered populations have long been recognized. However, the correspondence between engineered and native embryonic progenitors has been relatively less well characterized, largely due to their transient and highly heterogenous nature, and therefore many studies have relied on expression of a few canonical markers to ensure that their cells are on the correct path. To address this challenge, we first generated an in vivo gastrulation mouse single cell RNA sequencing (scRNA-seq) reference data set and used it to train a collection of computational tools for comparing cell types, states, regulators, and trajectories across studies. Then we used these tools to evaluate a commonly used protocol designed to promote mesoderm derivation, as well as four previously published biomimetic protocols. Finally, we diversified our toolkits to reach a broader scientific community by implementing our primary analysis tool in Python and as an interactive web application.