Fig. 1.
An illustration of the PseudotimeDE method (Created with BioRender.com). The core of PseudotimeDE is to obtain a valid null distribution of the DE gene test statistic Sj for each gene j. To achieve that, PseudotimeDE subsamples 80% cells from the original scRNA-seq data. Then on each subsample, PseudotimeDE performs pseudotime inference (using a user-specified method such as Slingshot and Monocle3-PI) and permutes the inferred pseudotime across cells. Next, PseudotimeDE fits a model (NB-GAM or ZINB-GAM) to the permuted subsamples to obtain the values of Sj under the null hypothesis and uses these values to approximate the null distribution of Sj. In parallel, PseudotimeDE fits the same model to the original dataset and calculate the observed value of Sj. Finally, PseudotimeDE derives the p-value from the observed value and the null distribution of Sj. Detail is described in the “Methods” section