Extended Data Fig. 8: Benchmarking against scVelo.

a-b, Representative test datasets comparing CytoTRACE 2 and scVelo. a, UMAP representation of mouse pancreas development (10x) (Supplementary Table 1). Left: Cells colored by ground truth granular potency level (Fig. 1b; Supplementary Table 3). Center: Cells colored by CytoTRACE 2 potency scores. Right: Cells colored by scVelo latent time (differential kinetics model). b, Same as a but showing human bone marrow (CITE-seq) (Supplementary Table 1). c, Left: Bar plot showing mean absolute order (weighted τ applied to single cells) performance across six broad potency levels (circles) and ten granular order potency levels (triangles) for nine test datasets evaluable by CytoTRACE 2 and scVelo (Supplementary Tables 3 and 14; Methods). Two models are shown for the latter: dynamical latent time and differential kinetics latent time. Right: Violin and box plots showing relative order performance (weighted τ applied to single cells) on the same test datasets (n = 8 evaluable datasets with relative developmental orderings, Supplementary Tables 4 and 14). Statistical significance was determined by two-sided paired t test. Violin plot bounds denote minimum and maximum values. Box center lines, bounds of the box, and whiskers denote medians, 1st and 3rd quartiles, and minimum and maximum values within 1.5 × IQR (interquartile range) of the box limits, respectively. d, Same as Fig. 1e but shown for the nine evaluable test datasets in c. Left: CytoTRACE 2 potency scores. Right: scVelo latent time (differential kinetics model). Statistical significance was determined using a one-sided Z-test. ns, not significant.