Question 3 of 6 Intermediate Concept
Why should donors rather than individual cells determine biological replication?
Short answer
Cells from one donor share genetics, environment and sample processing, so they are not independent biological replicates. For questions about treatment or disease across people, inference should generally use donors or independent samples, not thousands of cells as the sample size.
Why
Suppose five patients contribute 10,000 T cells each. The experiment has five independent patient units, not 50,000 biological replicates. Treating cells as independent can drastically understate uncertainty and inflate false discoveries.
A widely used approach sums counts by donor, condition and annotated cell type to create pseudobulk profiles, then fits a donor-aware count model. Mixed-effect approaches may also be appropriate for hierarchical data. The correct experimental unit depends on how treatments were assigned and samples collected, not merely the fact that data were collected per cell.
A donor with thousands of cells may provide precise information about that donor, but not substitute for a second person. Uneven cell numbers, compositional shifts and treatment-linked capture differences require careful handling. State the target estimand and explain whether the analysis concerns changes in average gene expression within a cell type, cell-type proportions, or individual-cell variation.
Common mistakes
- Claiming that 50,000 cells from five donors confer the statistical power of 50,000 independent patients
The interviewer might ask next
When should you use pseudobulk, and when might a hierarchical model be preferable?