Pseudotime Trajectory Inference

Employs reversed graph embedding to learn a principal graph that describes the trajectory. It first performs UMAP dimensionality reduction, then uses partition-based graph abstraction (PAGA) to construct a coarse-grained trajectory graph. Cells are ordered along this graph using geodesic distances, enabling the reconstruction of complex topologies including branching points and convergent differentiation. Monocle 3 scales to millions of cells and supports trajectory-resolved differential expression analysis.

A top-performing method that infers lineage structures and pseudotime simultaneously. Slingshot identifies the global lineage structure by fitting a minimum spanning tree (MST) through cluster centers in a reduced-dimensional space, then refines each lineage by fitting simultaneous principal curves. This two-stage approach handles multiple branching events and produces smooth, robust pseudotime assignments without requiring a starting cell to be specified a priori.

A fundamentally different approach that predicts the future state of individual cells by modeling the ratio of unspliced precursor mRNA to mature spliced mRNA. The difference between these two quantities—the RNA velocity vector—indicates whether a gene is being up- or down-regulated. Algorithms like scVelo extend this with a likelihood-based dynamical model that recovers the full transcriptional kinetics of splicing, enabling the inference of latent time and directional flow without requiring trajectory graph construction.

Generates a coarse-grained connectivity graph that preserves the global topology of the data while abstracting away fine-grained structure. PAGA estimates connectivity between clusters by computing a statistical measure of inter-cluster edge density in the k-nearest neighbor graph. The resulting graph reveals disconnected lineages, branching, and convergent differentiation paths. It is often used as a preprocessing step for other pseudotime methods or as a standalone interpretable summary of developmental topology.

A method based on diffusion maps, a nonlinear dimensionality reduction technique that models the transition probabilities of a random walk on the cell-cell similarity graph. DPT defines pseudotime as the diffusion distance from a user-specified root cell to all other cells. This distance metric is robust to noise because it integrates over all possible paths through the data, not just the shortest one. DPT naturally handles bifurcations and provides a measure of branching confidence.

Models cell fate decisions as a Markov chain on the neighborhood graph. Starting from an early cell, Palantir computes absorption probabilities toward terminal states using an iterative random walk. This yields a continuous pseudotime ordering and, critically, a branching probability vector for each cell that quantifies the likelihood of adopting each terminal fate. This probabilistic framework captures stochastic cell fate commitment and identifies cells in transitional, low-commitment states.