Fig. 3: Velocity model comparison in complex biological systems.

a, Comparison of velocity estimation when using different algorithms to quantify unspliced and spliced counts. Correlation of velocities derived from pairs of quantification algorithms and from velocities estimating using one of veloVI (VI), the EM and steady-state (SS) model are compared on the pancreas³⁵ and spermatogenesis³⁶ data. Box plots indicate the median (center line), interquartile range (hinges) and whiskers at 1.5× interquartile range (n = 78 pairs of quantification methods each). b, Comparison of veloVI and the EM model based on gene-wise MSE (left), cell-wise velocity consistency (middle) and gene-wise velocity Pearson correlation (right) on datasets of pancreas endocrinogenesis³⁵ (n = 1,074 genes for MSE and velocity correlation; n = 3,696 cells for velocity consistency), hippocampus⁸ (n = 1,292 genes for MSE and velocity correlation; n = 18,213 for velocity consistency), forebrain⁸ (n = 822 genes for MSE and velocity correlation; n = 1,720 for velocity consistency) and retina⁴³ (n = 700 genes for MSE and velocity correlation; n = 2,726 for velocity consistency). Box plots indicate the median (center line), interquartile range (hinges) and whiskers at 1.5× interquartile range. c, Velocity comparison on the level of individual genes in the pancreas dataset (Sulf2 and Top2a). For each gene, the velocity of the EM model is plotted against veloVI (left) and the gene phase portrait is given (right). Each observation is colored by its cell type as defined in previous work³⁵.