Questions: Single-Cell RNA Sequencing

Question 1

Why is dropout (failure to detect expressed genes) a more significant problem in scRNA-seq than in bulk RNA-seq?

• A. scRNA-seq uses lower quality sequencing chemistry
• B. Each cell provides only picograms of RNA, so low-abundance transcripts are frequently lost during capture and amplification
• C. scRNA-seq does not perform PCR amplification
• D. Bulk RNA-seq uses longer reads that capture more genes

Answer: (B)

A single cell contains roughly 10 picograms of total RNA — orders of magnitude less than a bulk sample. During library preparation, the capture and reverse transcription of such small quantities is inherently stochastic: a transcript present at a few copies may be captured in one cell but missed in another. This produces an excess of zero counts (dropout) that does not reflect true absence of expression. Specialized imputation methods and models that account for zero-inflation are used to handle this.

Question 2

Cell clusters identified in a UMAP visualization of scRNA-seq data always correspond to distinct, discrete cell types.

• T. True
• F. False

Answer: False

UMAP is a dimensionality reduction method that preserves local neighborhood structure, but it can create apparent clusters from continuous data and can separate or merge groups depending on parameter settings. Apparent clusters may represent discrete cell types, continuous differentiation states, cell cycle phases, or even technical batch effects. Biological validation using known marker genes, trajectory analysis, and independent experimental methods is needed to determine whether clusters represent genuinely distinct cell populations.

Question 3

Explain the role of highly variable gene (HVG) selection in the scRNA-seq analysis workflow.

Model answer: HVG selection identifies genes whose expression varies more across cells than expected from technical noise alone. These genes carry the biological signal that distinguishes cell types and states. By restricting downstream analysis (PCA, clustering) to the top 1,000-3,000 HVGs, you remove the noise contributed by non-informative genes (housekeeping genes, lowly expressed genes dominated by dropout) and focus computational resources on the features that actually differentiate cells. This improves clustering quality and reduces computational cost.

Without HVG selection, PCA would be dominated by technical noise and highly expressed but non-variable housekeeping genes, producing components that do not separate cell types. The variance-mean relationship is used to identify genes with excess variability: genes must have variance above what is expected at their mean expression level to qualify as highly variable.