Skip to main content
Fig. 3 | Genome Biology

Fig. 3

From: PseudotimeDE: inference of differential gene expression along cell pseudotime with well-calibrated p-values from single-cell RNA sequencing data

Fig. 3

PseudotimeDE outperforms four state-of-the-art methods (tradeSeq, Monocle3-DE, NBAMSeq, and ImpulseDE2) for identifying DE genes along cell pseudotime. Left panels a–e are based on pseudotime inferred by Slingshot; right panels f–j are based on pseudotime inferred by Monocle3-PI. a, f Distributions of non-DE genes’ observed p-values by five DE methods with inferred pseudotime. Top: quantile-quantile plots that compare the empirical quantiles of the observed p-values against the expected quantiles of the Uniform [0,1] distribution. Bottom: histograms of the observed p-values. The p-values shown on top of histograms are from the Kolmogorov–Smirnov test under the null hypothesis that the distribution is Uniform [0,1]. The larger the p-value, the more uniform the distribution is. Among the five DE methods, PseudotimeDE’s observed p-values follow most closely the expected Uniform [0,1] distribution. b, g Quantile-quantile plots of the same p-values as in a and f on the negative log10 scale. PseudotimeDE returns better-calibrated small p-values than the other four methods do. c, h FDPs of the five DE methods with the target FDR 0.05 (BH adjusted- p≤0.05). PseudotimeDE yields the FDP closest to 0.05. d, i ROC curves and AUROC values of the five DE methods. PseudotimeDE achieves the highest AUROC. e, j Power of the five DE methods under the FDP = 0.05 cutoff. PseudotimeDE achieves the highest power

Back to article page