Discovering novel cell types across heterogeneous single-cell experiments [article]

Maria Brbic, Marinka Zitnik, Sheng Wang, Angela O. Pisco, Russ B. Altman, Spyros Darmanis, Jure Leskovec
2020 bioRxiv   pre-print
Although tremendous efforts have been put into cell type classification, the identification of previously uncharacterized cell types in heterogeneous single-cell RNA-seq data remains a challenge. We present MARS, a meta-learning algorithm for annotating known and novel cell types that reconciles the heterogeneity by transferring latent cell representations across multiple datasets. MARS jointly learns a set of cell-type landmarks and a cell embedding space in which cells group by landmark while
more » ... landmarks of distinct cell types are embedded far away from each other. The method annotates cells by probabilistically defining a cell type based on landmark positions in the embedding space. MARS has a unique ability to discover cell types that have never been seen before and characterize experiments that are yet unannotated. We apply MARS to a large aging cell atlas of 23 tissues covering the life span of a mouse. MARS accurately identifies cell types, even when tissues have no cell types in common. Further, the method automatically generates interpretable names for novel cell types. Remarkably, MARS estimates meaningful cell-type-specific signatures of aging and visualizes them as trajectories reflecting temporal relationships of cells in a tissue.
doi:10.1101/2020.02.25.960302 fatcat:3ecinom22ffztjwqmtlndyn76i