Trajectory and Developmental Analysis
When cells are in motion
Cells change in response to stimuli, differentiate, progress through disease states and can undergo dramatic phenotypic shifts over time. Clustering captures a snapshot of the cell types and cell states at a single moment. Trajectory inference reconstructs the continuous paths cells travel as they transition from one state to another, revealing the sequence of molecular events and the branch points where fate decisions occur.
This matters whenever the biology you care about is dynamic rather than static, e.g., progenitor cell differentiation, wound healing responses, cancer progression, immune cell activation, or drug response. If your experiment was designed to capture a process rather than a stable cell type, trajectory analysis is the right framework for interpreting the data.
Pseudotime and trajectory inference
Pseudotime orders cells along a continuous axis that reflects their progress through a biological process. Monocle 3 and Slingshot are the most widely used tools for this analysis, each offering different approaches to topology estimation (graph-based vs. principal curves). The choice between them depends on whether the trajectory is expected to have a single lineage, multiple branches, or a convergent structure.
For experiments with time series data, supervised pseudotime tools such as psupertime can incorporate known collection times to improve inference. Cross-validation of pseudotime with RNA velocity (see below) is good practice, as it tests whether the inferred ordering is consistent with the kinetic information in the unspliced/spliced ratio.
RNA velocity
RNA velocity uses the ratio of unspliced to spliced mRNA as a proxy for the direction and rate of transcriptional change in each cell. Because newly transcribed pre-mRNA is captured alongside mature mRNA in most single-cell protocols, it is possible to estimate whether a gene is being actively induced or suppressed, and therefore to predict the short-term trajectory of each cell independently of pseudotime assignments.
scVelo remains widely used and is integrated into the Scanpy/AnnData ecosystem. VeloVI (a variational inference extension published in Nature Methods, 2023) is preferred for noisy datasets where the steady-state assumptions of the base model break down. VeloTrace (2026 preprint) represents an emerging approach that reconciles velocity and trajectory into a unified framework using neural ordinary differential equations, offering improved accuracy on complex topologies. For most production work, scVelo with VeloVI fallback is the current standard.
Spatial trajectory analysis
When single-cell trajectory data is paired with spatial transcriptomics, it becomes possible to ask where in the tissue different stages of a developmental or disease process are occurring. Spatially resolved trajectory analysis maps pseudotime gradients onto physical tissue architecture, connecting molecular transitions to anatomical location.
This is particularly informative in tissues with defined anatomical gradients, for example intestinal crypt-villus axes, skin epidermal differentiation from basal to superficial layers, or tumor invasion fronts. In these settings, spatial trajectory analysis converts qualitative anatomical descriptions into quantitative transcriptional measurements.