PseudoVelo: Inferring Gene Expression Derivatives Along Pseudotime as Pseudo-Velocity

Understanding single-cell transcriptional dynamics during cellular differentiation and transition is fundamental to developmental biology. RNA velocity serves as a valuable approach for inferring these dynamics but is constrained by its reliance on simplified splicing kinetics. As a kinetics-free alternative, pseudotime-based approaches have been developed to reconstruct cellular transitions. Nevertheless, these approaches merely estimate cell-cell transitions by biasing the edges of a nearest-neighbor graph toward mature cell states. Here, we present PseudoVelo, a computational method that infers gene expression derivatives along pseudotime as "pseudo-velocity". Utilizing the Generalized Additive Models commonly applied in pseudotime analysis, PseudoVelo fits the expression of each gene as a function of pseudotime. Then, by employing a central difference approximation, our method directly calculates the derivative of gene expression with respect to pseudotime, thereby obtaining the pseudo-velocity for each individual cell. Evaluated on multiple developmental processes, PseudoVelo demonstrates strong performance compared to CellRank 2, effectively recovering correct cellular trajectories using diverse temporal priors and demonstrating high resilience against various data perturbations.