Unify learns cellular evolution with universal multimodal embeddings

Unify is a transfer learning method that integrates single-cell RNA-seq data across species by creating universal cell embeddings using multi-modal macrogenes combining RNA expression and protein/language model embeddings. It enables cross-species comparisons beyond gene orthology, corrects batch effects, predicts perturbation responses, and reconstructs evolutionary cell-type trees across species separated by over 700 million years.