Deep learning-based models for preimplantation mouse and human embryos based on single-cell RNA sequencing.

Area of Science:

• Developmental Biology
• Computational Biology
• Genomics

Background:

• Single-cell transcriptomics generates vast datasets for embryonic development and stem cell models.
• Defining cell types and states in vivo and comparing them to in vitro differentiation is challenging due to data complexity.

Purpose of the Study:

• To develop and apply deep learning tools for integrating and classifying multiple single-cell transcriptomic datasets.
• To define mouse and human embryonic cell types, lineages, and states from challenging-to-obtain early developmental samples.
• To create an unbiased classification model and identify key genes for lineage and cell-type identification.

Main Methods:

• Utilized deep learning algorithms to integrate and classify diverse single-cell transcriptomic datasets.
• Developed a classification model trained on in vivo embryonic development data.
• Applied the trained model to classify pluripotent stem cell-derived models for both mouse and human development.

Main Results:

• Successfully defined mouse and human embryonic cell types, lineages, and states.
• Identified the specific gene sets utilized by the deep learning model for classification.
• Demonstrated the utility of the model in classifying in vitro stem cell differentiation, aligning with in vivo development.

Conclusions:

• Deep learning provides a powerful approach to integrate and analyze complex single-cell transcriptomic data.
• The developed model serves as a dynamic reference for early embryogenesis, aiding in the interpretation of both in vivo and in vitro developmental processes.
• This work maximizes insights from precious early developmental samples and facilitates comparisons between natural development and stem cell-based models.