Fig. 2: TIVelo’s efficacy and the comparison with ODE-based method.
From: TIVelo: RNA velocity estimation leveraging cluster-level trajectory inference
a Gene Mllt3 in mouse gastrulation (erythroid) fitted by ODE curves of scVelo. Mllt3 is a gene with top likelihood fitted by scVelo. b Gene Gclm in mouse gastrulation (erythroid) fitted by ODE curves of scVelo. Gclm is a multiple rate kinetics gene (MURK gene), with rate parameter α suddenly increasing at some time point25,28 (top left panel). c RNA velocity stream plot inferred from scVelo, based on genes with top 500 likelihood except MURK genes. d RNA velocity stream plot inferred from scVelo, based on MURK genes. e The u-s pattern along the main path for corresponding two genes. In each panel the colorbar below indicates the cell type, which is identical to the annotation in (a). f The velocity graph for selected cells created by scVelo. There is a unit length arrow showing the direction from selected cell n to cell \(n^{\prime}\) on the UMAP embedding plot if \(n^{\prime}={{{\mathrm{argmax}}}\,}_{n^{\prime} }{\pi }_{n,n^{\prime} }\). The line width of each arrow is proportional to its corresponding \({\pi }_{n,n^{\prime} }\). g RNA velocity stream plot produced by scVelo. h TIVelo first infers the direction on the cluster level, according to the orientation scores. i In RNA velocity estimation, for each cell and each gene, the inferred velocity should follow the direction inferred in (h). j The velocity graph for selected cells created by TIVelo. k RNA velocity stream plot produced by TIVelo.
