TemporalVAE for Temporal Prediction on Small Datasets v1

TemporalVAE is a deep generative model in a dual-objective setting to infer the biological time of cells from a compressed latent space. TemporalVAE exhibits scalability to millions of cells in the mouse development atlas and high accuracy in atlas-based cell staging on mouse organogenesis across platforms, as well as during human peri-implantation between in vivo and in vitro conditions. Here we describe the workflow of TemporalVAE, including predicting the biological time of samples and identifying temporally sensitive genes.