Fig. 3
From: Modeling group heteroscedasticity in single-cell RNA-seq pseudo-bulk data
Group variance modeling methods provide good power while controlling the false discovery rate. a Mean-variance trends plotted for each group by the voomByGroup function on simulated scRNA-seq data with varying group-specific common BCV values, where colors represent different groups. In terms of the simulated variability-level, groups 1 and 2 represent Low variation, while 3 and 4 have High variation. Based on DE analysis results, FDR across methods in different comparisons (colors denote the 3 comparison types) are summarized in panel (b) at a cut-off of 0.05. The number of DE genes recovered by different methods for comparison between the two groups with higher variability (panel c) and the two groups with lower variability (panel d) at the same FDR cut-off are shown. For each bar in these plots, gray represents true positive genes, red represents false positive genes, and the FDR is labeled at the top of the bar