Fig. 1: NK cell differentiation at the transcriptional level.
a, Integration process of scRNA-seq data of NK cells from 12 donors and 4 different laboratories using scVI showing a UMAP based on the scVI latent representation, followed by a UMAP based on the diffusion map components. b, AUCell scores of gene signatures for CD56bright and CD56dim NK cell subsets. c, UMAP representation of five sorted subsets from a donor with an adaptive expansion (left) and a donor without an adaptive expansion (right). d, Heatmap depicting accuracy of our prediction model for subset annotation tested on the held-out 15% of cells from the subset-specific dataset (two donors). e, UMAP of the scANVI representation of both bulk and sorted NK cells, showing original annotation of NK cells (12 donors, left) and subset labels predicted (right) using the scANVI model trained with sorted subset data (2 donors). f, Dot plots showing the top three up- and downregulated genes between all pairs of subsets (x and y axes) as identified by the differential expression module in scANVI. These top genes were then visualized across all NK cell subsets within the differentiation spectrum (x axis), to highlight the continuous nature of NK cell differentiation. g, Diffusion map representation showing the predicted subset labels for the bulk data (top) and depicting Leiden clustering of the 12 donor NK cell dataset (bottom). h, Heatmap showing distribution of our annotated 12 donor NK cell subsets over the 5 Leiden clusters. i, Frequency (freq.) of annotated late CD56dim and adaptive NK cell subsets in donors with and without an adaptive NK cell expansion. Int., intermediate.
