Stem cell-derived embryo models offer a powerful tool for studying early development, but variability in their formation challenges research standardization. The doctoral candidate Luca Deininger and a team at Caltech have applied deep learning to enhance the reproducibility of selecting ETiX-embryos, classifying 900 post-implantation structures into normal and abnormal categories. Using live imaging and AI-based models, they achieved 88% accuracy at 90 hours post-cell seeding and 65% accuracy at the initial stage, predicting developmental trajectories. The study identifies key features of normal ETiX-embryos, such as higher cell counts, larger size, and compact morphology, and confirms through perturbation experiments that increasing initial cell numbers improves development outcomes.

This work was funded by the Helmholtz Association as part of the graduate school “HIDSS4Health” (Helmholtz Information and Data Science School for Health) and the Helmholtz program NACIP (Natural, Artificial and Cognitive Information Processing), each covering 50% of the personnel and travel costs. The High Performance Computing (HPC) on the HAICORE@KIT partition was supported by the Helmholtz Association Initiative and Networking Fund.

The team reported their findings in the journal Nature Communications (https://doi.org/10.1038/s41467-025-56908-5).