Framework Benchmarks Lab-Grown Embryo Models Against Real Development

Researchers at the University of Sydney have built a computational framework that measures how accurately lab-grown embryo models reflect real human embryonic development, revealing that while certain models perform well, none fully capture the complexity of the earliest stages of life.

These models, called blastoids, are generated from stem cells and give scientists a way to study the biology of fertility, pregnancy success, and early development without depending on donated human embryos. They are research models rather than actual embryos, and current versions cannot develop into one. Stem-cell-derived models offer a path around questions that have long been hard to study due to technical and ethical limits, but resembling an embryo in appearance does not necessarily mean behaving like one biologically. 

The Sydney team addressed that gap by evaluating four leading human blastoid-generation methods and uncovering substantial differences in how well each reproduces the cell types and developmental processes found in natural human embryos. The study, published in Cell Systems, assembled one of the most comprehensive reference maps of early human embryo development and used it to benchmark the biological accuracy of these stem-cell-derived models.

Lead author Pengyi Yang said the work gives researchers a more objective way to weigh the strengths and limitations of embryo models. “Human embryo models have enormous potential for studying the earliest days of an embryo’s development, but there has been no consistent way to assess how accurately these reflect real human development,” he said. “Our framework allows researchers to compare these models against a detailed biological reference and determine which cell types and developmental processes are faithfully reproduced, and which are not.”

To build the reference map, the team combined and harmonized more than 14,000 single-cell transcriptomes from human embryos across key developmental stages before and after implantation, then compared that map against four widely used blastoid protocols from international groups. Rather than judging resemblance under a microscope, they examined molecular identity, developmental timing, and lineage structure. Some models reproduced all three major lineages of a natural blastocyst relatively well; others missed certain cell types or contained many cells unmatched to any known embryonic state, and no model replicated a natural blastocyst perfectly.

Yang said the findings show both promise and limits: “The encouraging finding is that some models capture important aspects of early embryonic development relatively well, although each model has limitations. But our study also shows that current models are not biologically equivalent to real human embryos, and researchers need to be careful about the conclusions they draw from them.” He added, “If we’re going to use these systems to answer important biological questions, we first need to know what they can reliably tell us. Our work provides a roadmap for improving embryo models and ensuring scientific claims remain grounded in what the models can actually support.” The team has made its reference datasets and benchmarking tools publicly available for researchers worldwide.