Cell pseudotime conversation score

We inferred the pseudotime from the corrected data of each method using Monocle 2 [53]. The pseudotime consistency between the inferred pseudotime and the gold standard pseudotime (i.e., ground truth) was evaluated by two metrics: pseudotime ordering score (POS) and Kendall rank correlation coefficient. The POS is defined as [54]

where C represents the number of similar orders between the pseudotime and gold standard orders, and N_c denotes the number of dissimilar orders.

For the mouse skin embryonic development scRNA-seq dataset, the gold standard trajectories and pseudotime are lacking. To compare the recovered trajectories in a quantitative fashion, we use a similar method from Luecken et al. [19], which assumes that trajectories inferred using the unintegrated data for each batch are the more accurate biological signal. For the unintegrated data, we compute the pseudotime values of cells by analyzing each replicate sample separately and then calculate Pearson correlation coefficients between the pseudotime values of cells in each replicate sample before and after integration. To infer the pseudotime values of cells, we adopt a principal curve-based approach used in the previous pseudotime-inference tools such as scEpath [31] and Slingshot [55]. Specifically, we fit a principal curve in the PHATE space and obtain the pseudotime values by orthogonal projections of cells onto the curves.