Fig. 2

Selection of variable genes. In the first two panels, each dot shows the variance of a single gene in the PBMC dataset after applying a normalization method. The dotted horizontal line shows a threshold adjusted to select 100 most variable genes. Red dots mark 100 genes that are selected by the other method, i.e., that are above the threshold in the other panel. Stars indicate genes shown in the last panel. a Gene variance after sequencing depth normalization, median-scaling, and the square-root transformation. b Variance of Pearson residuals (assuming θ=100). c t-SNE of the entire PBMC dataset (see Additional file 1: Figure S4), colored by expression of four example genes (after sequencing depth normalization and square-root transform). Platelet marker Tubb1 with low average expression is only selected by Pearson residuals. Arrows indicate the platelet cluster. Fos is only selected by the square root-based method, and does not show a clear type-specific expression pattern. Malat1 (expressed everywhere apart from platelets) and monocyte marker Lyz with higher average expression are selected by both methods