Fig. 5

Evaluation of DE analysis methods in single-cell expression data sets in which cells were simulated from two groups without partial membership to these groups. A and B compare posterior mean LFC estimates and posterior z-scores returned by DESeq2 [79] and GoM DE. Each plot shows 200,000 points for 10,000 genes \(\times\) 20 simulated data sets. C summarizes performance in identifying differentially expressed genes in all simulated data sets; it plots power and false discovery rates (FDR) for the three methods compared as the p-value (MAST [83]), s-value (DESeq2), or lfsr threshold (GoM DE) is varied from 0 to 1. Power and FDR are calculated from the number of true positives (TP), false positives (FP), true negatives (TN), and false negatives (FN) as FDR = FP/(TP + FP) and power = TP/(TP + FN). See also Additional file 1: Figs. S2, S3