Questions: Single-Cell Trajectory Analysis

Question 1

Pseudotime analysis of a scRNA-seq dataset places cells along a trajectory from stem cells to differentiated cells. What does the pseudotime value represent?

• A. The actual chronological time since the cell was born
• B. An inferred ordering that reflects the relative progress of each cell through the biological process, based on gene expression similarity — cells with similar pseudotime values have similar transcriptional states, but the mapping to real time is generally nonlinear and unknown
• C. The number of cell divisions the cell has undergone
• D. The distance of the cell from the centroid of the dataset in PCA space

Answer: (B)

Pseudotime is an abstract coordinate that orders cells by transcriptional similarity along a trajectory, not a measurement of real time. Two cells with pseudotime values of 0.3 and 0.6 are inferred to be at different stages of the process, with 0.6 being further along, but this does not mean 0.6 has been differentiating for twice as long. The relationship between pseudotime and real time depends on the rate of transcriptional change, which may vary across the trajectory — rapid transitions compress pseudotime, slow transitions stretch it. Calibrating pseudotime to real time requires additional information (time-stamped samples, pulse-chase experiments, or live imaging).

Question 2

RNA velocity determines the direction of cell-state transitions by comparing future and past gene expression measurements from the same cell.

• T. True
• F. False

Answer: False

RNA velocity does not require measuring the same cell at two time points — that would be impossible since scRNA-seq is destructive. Instead, it exploits a snapshot measurement: within each cell, the ratio of unspliced pre-mRNA to spliced mature mRNA reflects whether a gene is being upregulated (unspliced > expected at steady state, meaning transcription recently increased) or downregulated (unspliced < expected, meaning transcription has decreased but mature mRNA persists). By computing this ratio across all genes, RNA velocity infers a vector in gene expression space that points toward the cell's predicted future state. The key biological insight is that unspliced and spliced mRNA act as a natural 'time derivative' built into every cell — the unspliced fraction is a leading indicator of where expression is heading.

Question 3

What is the fundamental assumption that allows trajectory inference from a single-time-point scRNA-seq experiment, and when does this assumption fail?

Model answer: The fundamental assumption is ergodicity (or quasi-ergodicity): that the population of cells captured at a single time point contains cells at all stages of the biological process, so the distribution of cell states in the sample recapitulates the temporal progression. This assumes the process is asynchronous — not all cells started at the same time — so that cells spanning the full trajectory are present simultaneously. This assumption fails when the process is highly synchronized (all cells transition together, as in stimulus-response experiments where the entire population is perturbed simultaneously), when rare intermediate states are undersampled, or when the process has not yet begun or has fully completed at the time of capture. In synchronized processes, all cells are at approximately the same stage, providing no information about the trajectory.

The ergodic assumption is borrowed from statistical mechanics: measuring many particles at one time point is equivalent to measuring one particle at many time points, if the system is in a steady state of flow. In biology, this works well for continuous developmental processes (hematopoiesis, spermatogenesis) where cells constantly enter and progress through the trajectory, but fails for acute responses (all cells responding to a drug simultaneously).

Question 4

How does Slingshot differ from Monocle 3 in its approach to trajectory inference?

Model answer: Slingshot and Monocle 3 differ primarily in their dimensionality reduction and curve-fitting strategies. Slingshot operates in a reduced-dimension space (typically from PCA), constructs a minimum spanning tree (MST) on cluster centroids to identify the global trajectory topology (linear, branching), then fits simultaneous principal curves through the data along each lineage path, producing smooth pseudotime orderings. Monocle 3 uses UMAP for dimensionality reduction, learns a principal graph (not just an MST on clusters but a graph learned directly from the data points) in the UMAP space, and assigns pseudotime by geodesic distance along this graph from a user-specified root. Monocle 3 handles more complex topologies (loops, convergences) through its graph-learning approach, while Slingshot's principal curve fitting tends to produce smoother trajectories for simpler branching structures.

Both methods require the user to specify a root cell or starting point, which introduces subjectivity. In practice, the choice of method matters less than the biological validation of the resulting trajectory — do known marker genes change monotonically along pseudotime? Do branch points correspond to known fate decisions? Computational trajectory inference generates hypotheses that must be validated by independent experimental methods (lineage tracing, time-course experiments).