Question 19 of 20 Advanced Concept
What can RNA velocity tell you, and which assumptions limit it?
Short answer
RNA velocity compares unspliced and spliced counts per gene to infer whether a gene is being switched on or off, and combines genes to suggest where each cell is heading. It assumes simple kinetics with constant rates, so it can point arrows in arbitrary directions where those assumptions fail.
Why
A newly induced gene accumulates unspliced pre-mRNA before mature mRNA catches up, and a gene being shut down shows the reverse. The Bioconductor trajectory chapter explains that a high ratio of unspliced to spliced reads suggests upregulation, and a low ratio suggests downregulation, which can give a trajectory a direction. Unspliced counts usually come from intronic reads.
The review by Bergen, Soldatov, Kharchenko and Theis sets out the assumptions. Current models treat each gene independently with first-order kinetics and constant rates. The original steady-state model also assumes a common splicing rate across genes and that steady states are at least partly observed in the data.
The dynamical model in scVelo relaxes the steady-state assumption but keeps constant rates. When rates change, for example through a transcriptional boost during erythroid maturation, curvature in the data can be inflated, deflated or flipped. The review shows mature blood cells with no real dynamics given arbitrary directions.
Velocity is best read as a hypothesis about direction, checked against known biology, time points or lineage tracing.
A strong answer covers
- Unspliced to spliced ratio indicates induction or repression
- Genes are modeled independently with constant rates
- Steady-state model assumes a common splicing rate
- Changing kinetic rates can distort or flip inferred direction
- Mature, non-dynamic populations can receive arbitrary arrows
Common mistakes
- Treating velocity arrows on a UMAP as proof of lineage
- Applying velocity to steady-state mature populations and reading the arrows
- Ignoring how intronic counting choices affect the unspliced matrix
The interviewer might ask next
What experimental data would you use to validate a direction inferred by RNA velocity?