Fig. 1

Icebear’s cross-species prediction and comparison framework. A Icebear is trained on single-cell RNA-seq profiles across species. Specifically, Icebear uses conditional variational autoencoders that predict cellular profiles from a combination of species and batch factors, as well as species-invariant cell factors. B Once the model is trained, we can swap the species factor to predict a (e.g., mouse) cell’s corresponding profile in another species (e.g., human). C Icebear can be used to predict single-cell profiles in species (e.g., human) where the corresponding cell type has not been profiled. D Icebear can also perform cross-species comparison at the single-cell level, revealing X chromosome upregulation patterns during evolution. Components of this figure were created with Biorender