Single-Cell Trajectory Analysis

Core Idea

Single-cell trajectory analysis reconstructs the continuous progression of cells through biological processes — differentiation, activation, disease progression — from snapshot scRNA-seq data where each cell is measured only once. Since single-cell RNA sequencing destroys the cell, temporal ordering must be inferred computationally: cells at different stages of a process coexist in the sample, and trajectory inference algorithms arrange them along a pseudotime axis that recapitulates the biological progression. Pseudotime methods (Monocle, Slingshot, PAGA) construct low-dimensional manifolds from gene expression space and order cells along paths through these manifolds. RNA velocity (La Manno et al., 2018) adds directionality by exploiting the ratio of unspliced to spliced mRNA within each cell as a proxy for transcriptional rate of change, predicting each cell's future state without requiring external time labels. Together, these methods transform static snapshots into dynamic narratives of cell-state change, making trajectory analysis central to modern developmental and stem cell biology.

Explainer

Single-cell RNA sequencing captures the transcriptomes of thousands to millions of individual cells, revealing the full heterogeneity of cell states within a tissue. But scRNA-seq provides only a snapshot — each cell is measured once and destroyed. If you want to understand a dynamic process like differentiation (how a stem cell becomes a neuron or a blood cell), you face a fundamental problem: you cannot follow individual cells through time. Trajectory inference solves this by exploiting the fact that in most biological processes, cells are asynchronous — at any given moment, cells at different stages of the process coexist in the tissue. By computationally ordering these cells by their transcriptional similarity, you can reconstruct the trajectory that any individual cell would follow over time.

Pseudotime methods — including Monocle (Trapnell et al., 2014), Slingshot (Street et al., 2018), and PAGA (Wolf et al., 2019) — construct this ordering algorithmically. The general approach is: (1) reduce the high-dimensional gene expression matrix to a lower-dimensional representation (PCA, diffusion maps, UMAP), (2) identify the topology of the trajectory (linear, branching, cyclical) using graph-based methods, and (3) assign each cell a pseudotime value reflecting its position along the trajectory. The result is a continuous ordering from progenitor states to differentiated states, along which you can identify genes that are dynamically regulated, branch points where fate decisions occur, and transcription factors that drive transitions. The critical assumption is ergodicity — that the snapshot population samples all stages of the process — which holds well for ongoing processes like hematopoiesis but fails for synchronized acute responses.

RNA velocity (La Manno et al., 2018) added a transformative dimension to trajectory analysis: directionality. Standard pseudotime methods infer an ordering but cannot intrinsically determine which end is the beginning and which is the end without prior biological knowledge (the user must specify a root). RNA velocity solves this by exploiting a signal internal to each cell: the ratio of unspliced pre-mRNA to spliced mature mRNA. Under a simple kinetic model, a gene being actively upregulated has an excess of unspliced mRNA relative to the steady-state expectation (transcription has increased, but the new transcripts have not yet been spliced). A gene being downregulated has a deficit of unspliced mRNA (transcription has decreased, but spliced mRNA persists). By computing this ratio across all genes, each cell gets a velocity vector in gene expression space — a prediction of its future transcriptional state. Projecting these vectors onto the low-dimensional embedding reveals the flow of cell-state transitions, including the directionality of differentiation and the location of attractor states (stable cell types where velocity approaches zero).

The practical impact of trajectory analysis on systems biology is profound. It has revealed previously unknown intermediate cell states in differentiation, identified transcription factor cascades driving fate decisions, and uncovered bifurcation points where a single progenitor population splits into multiple lineages. Tools like scVelo (Bergen et al., 2020) extended RNA velocity with a dynamical model that estimates gene-specific kinetic parameters (transcription, splicing, and degradation rates), improving accuracy and enabling the recovery of latent time — a quantity closer to real biological time than pseudotime. The integration of trajectory analysis with perturbation data (CRISPR screens in single cells), spatial transcriptomics (adding tissue location to trajectory position), and multi-omics measurements (simultaneous chromatin accessibility and gene expression) is making it possible to construct comprehensive, mechanistic models of cell-state dynamics that connect regulatory network architecture to developmental outcomes.