Fig. 3

Joint analysis of cells from all conditions leads to more accurate clustering of cell types compared to independent analysis of individual conditions. a Scatterplot illustrating the quality of clustering of cell types within each condition from the Mann et al. benchmark. Each point represents one cell type in one condition, when the embedding is computed using either the original expression data (“expression”), the embedding dimensions of scAlign, or the embedding dimensions of an autoencoder with the same neural network architecture as scAlign. The y-axis represents classification accuracy, while the x-axis represents the silhouette coefficient. b Same as a, but for HeterogeneousBenchmark. c tSNE plots visualizing the embedding space of scAlign trained on both conditions and d an autoencoder trained on a single condition