Pseudotime

Pseudotime is computed by measuring the minimum spanning tree or shortest path through a k-nearest neighbor graph of cells. The algorithm identifies a root node and assigns each cell a distance along the trajectory. Cells with highly similar transcriptomes are placed adjacent to one another, creating a smooth continuum rather than discrete clusters. This ordering is purely computational and requires no prior knowledge of real time points.

The starting point of pseudotime critically shapes the biological interpretation. Root cells are typically selected based on:

• Prior biological knowledge (e.g., earliest developmental stage)
• High expression of stemness markers
• Unsupervised methods that identify cells with the least differentiated transcriptomic signature Incorrect root selection can invert the trajectory, making differentiated cells appear as progenitors.

Unlike linear processes, many biological trajectories involve bifurcation points where cells commit to distinct lineages. Pseudotime algorithms detect these branch points by identifying regions in the transcriptional manifold where cell paths diverge. Common methods like Monocle and Slingshot fit principal curves or minimum spanning trees that can split, capturing the moment a common progenitor gives rise to two specialized cell types.

Once cells are ordered in pseudotime, gene expression can be modeled as a smooth function of progression. This reveals switch-like activation of transcription factors and gradual repression of pluripotency genes. Key analytical outputs include:

• Heatmaps of differentially expressed genes along pseudotime
• Gene-switch plots identifying the exact pseudotime value where a gene turns on or off
• Branch-dependent expression showing genes specific to one lineage fate

Pseudotime can be augmented with RNA velocity, which uses the ratio of unspliced to spliced mRNA to predict a cell's future state. While pseudotime provides a static ordering based on similarity, RNA velocity adds a directional arrow indicating whether a cell is moving toward differentiation or dedifferentiation. Combining both resolves ambiguities in trajectories where transcriptional similarity alone cannot distinguish forward from reverse progression.

Multiple algorithms exist for pseudotime inference, each with distinct assumptions:

• Monocle (DDRTree): Uses reversed graph embedding to learn a principal tree
• Slingshot: Fits simultaneous principal curves to cluster-based lineage endpoints
• PAGA (Partition-based Graph Abstraction): Estimates connectivity between clusters at coarse resolution while preserving trajectory topology
• Diffusion Pseudotime (DPT): Computes random-walk-based distances from a root cell, robust to noise in the manifold

The critical step of designating the starting point of a biological process, from which pseudotime values are measured. Incorrect root selection inverts or scrambles the trajectory.

• Biological priors: The earliest developmental stage, cells with highest stemness markers, or the cluster with the lowest transcriptional diversity
• Data-driven methods: Selecting cells with the highest expression of known early marker genes (e.g., CD34 for hematopoietic stem cells) or the extremum of a diffusion component
• RNA velocity integration: Cells with velocity vectors pointing outward but few pointing inward are candidate roots
• Consequence of mis-specification: The entire pseudotime ordering becomes biologically meaningless, as differentiation is a directed process

Root selection transforms an undirected graph into a directed trajectory, anchoring pseudotime in biological reality.

Statistical identification of genes whose expression changes as a function of pseudotime, revealing the transcriptional programs that drive progression through a biological process.

• TradeSeq: Fits a generalized additive model (GAM) with a negative binomial noise model to each gene's expression along pseudotime, detecting genes with significant non-linear patterns
• Monocle's Moran's I test: Identifies genes with significant spatial autocorrelation along the trajectory
• Output categories:
  • Early-activated genes: High expression near the root, decreasing over pseudotime
  • Late-activated genes: Low initially, rising toward terminal states
  • Transient genes: Peaking at intermediate pseudotime, often at branch points
• Biological insight: These gene sets define the molecular chronology of differentiation, identifying candidate regulators of each phase

This analysis bridges pseudotime ordering to functional biology, pinpointing the genes that execute developmental programs.