Fig. 6

Biologically meaningful DE results for the 10x Genomics PBMC dataset. a Scatterplot of the first two t-SNE dimensions obtained from the first ten principal components. Cells are color-coded by clusters found using the SEURAT graph-based clustering method on the first ten principal components. Pseudo-color images on the right display normalized enrichment scores after gene set enrichment analysis for cell types related to CD4+ T cells (see “Methods”), for clustering based on b the first ten principal components and cW from ZINB-WaVE with K=20. For dimensionality reduction, ZINB-WaVE was fitted with X=1 _n , V=1 _J , K=20 for W (based on the Akaike information criterion), common dispersion, and ε=10¹². To compute the weights for differential expression analysis, ZINB-WaVE was fitted with intercept and cell-type covariate in X, V=1 _J , K=0 for W, common dispersion, and ε=10¹². Normalized enrichment scores for more cell types are shown in Additional file 1: Figure S17. PCA principal component analysis