(a) Number of genes detected in Smart-seq3 and Smart-seq2 as a function of sequence depth. The median detection of genes across cells were represented as dots and the lines indicate the first and third quartiles. Separate plots were generated for cells of different cell types, as indicated on top of each figure item. The respective sample-sizes for Smart-seq3 and Smart-seq2 are: B-cell (n = 366, n = 112 cells), CD4+ T-cell (n = 1,270, n = 356 cells), CD8+ T-cell (n = 665, n = 222 cells), HEK cell (n = 236, n = 62 cells), Monocyte (n = 200, n = 302 cells), NK-cell (n = 352, n = 152 cells). (b) UMAP visualizations of sequenced HCA sample cells on different scRNA-seq protocols (data from Mereu et al. 2019), colored according to the Louvain clustering performed independently on cells from each protocol. The same computational pipeline and parameters was used for these analyses as in Figure 3a, except requiring a depth of just 10,000 reads per cell. Please note, this analysis is not intended to be a thorough benchmarking of methods as the data has merely been scaled and not sub-sampled to account for differences in sequencing depths or cell numbers between protocols. Instead the full data per protocol (Quartz-seq2: n = 1,422 cells, CEL-seq2: n = 750, Smart-seq2: n = 1,160, 10x v2 n = 3,592, 10x v3 n = 6,175 cells) was analyzed and run through a standardized scRNA-seq analysis pipelines, revealing that the B-cells do not easily separate with these other methods.