Abstract: Human embryonic development, from peri-implantation to gastrulation, represents a critical period for understanding cell lineage and tissue generation. With the growing number of stem-cell-based embryo models designed to replicate this stage in vitro, the need for annotating and evaluating these models using a well-curated, integrated post-implantation cell reference has become pressing. In this study, we employed a systematic approach to construct comprehensive reference atlases for human and monkey embryos. To facilitate automated cell type recognition, we further coupled our human reference with label transfer algorithms, which effectively identify cell types in both spatial and single-cell transcriptomic data. Furthermore, we curated annotations for nine human embryo models and benchmarked them using a series of metrics. Our reference-based annotation and benchmarking frameworks demonstrated their exceptional utility in characterizing embryonic models, assessing their fidelity, and revealing the extent of lineage development synchrony in current in vitro embryo models.
