A population average can conceal very different cellular patterns. Single-cell measurements help examine that variation, but individual dots still come from a measurement process and a sampling design. The value lies in understanding the distribution, rather than assuming that more dots automatically make the biological conclusion stronger.
Look beyond the average response
| Population | Cell-level values | Mean |
|---|---|---|
| A: four cells | 50, 50, 50, 50 | 50 |
| B: four cells | 0, 0, 100, 100 | 50 |
The two fictional populations have the same mean but different distributions. One has a uniform measured value; the other splits into two groups. A single bar at 50 would hide that difference.
This is a mathematical illustration using arbitrary units. Real cell measurements require uncertainty and a biological definition of the variable before the separation can be interpreted.
Read whether a paper is asking about a shift in the whole distribution, an unusual subset or the proportion crossing a defined threshold. These questions can produce different summaries of the same observations.
If the result is a percentage of responding cells, find the response definition. A threshold should be justified by the assay and comparison, rather than treated as a universal boundary between active and inactive biology.
Distinguish cell differences from measurement variation
Brennecke and colleagues’ original single-cell RNA-sequencing study developed a way to distinguish biological variability from technical noise. Its low-input experiments illustrate why an observed difference between cells does not automatically establish a true expression difference.Brennecke and colleagues — Accounting for technical noise in single-cell RNA-seq experiments (opens in a new tab)
For transcript measurements, an observed zero must be read in that measurement context. It should not automatically be rewritten as proof that the cell contains no molecules of the transcript.
Look for quality criteria and how the authors evaluated low-information observations. If some cells were removed, check whether the reported distribution refers to all captured cells or only those retained for analysis.
The same reading habit applies to other single-cell assays: identify the measured signal, its background and the criteria defining a usable observation. A dot is a record produced by an assay, not an unqualified view of the cell.
Separate subgroup proportions from within-group changes
Consider two fictional cell types with average values of 20 and 80. An equal mixture has an average of 50. A mixture containing three quarters of the higher-value type averages 65, even if neither type changes its own average.
That example shows how population composition can change a pooled result. Read whether the study compares corresponding cell groups, their relative abundance or the combined population.
Check how those groups were defined and labelled. A cluster label is part of an analysis and interpretation; look for the evidence linking it to the claimed cell type or state.
If a new cluster appears after exposure, ask whether the evidence supports a new biological state, a changed mixture or another explanation. The visual separation alone does not settle that question.
Keep independent samples visible
Squair and colleagues’ original evaluation of single-cell differential-expression methods found that failing to account for variation between biological replicates can generate false discoveries. A large number of cells does not remove the need to represent that sample structure.Squair and colleagues — Confronting false discoveries in single-cell differential expression (opens in a new tab)
For example, 2,000 cells from one donor are not 2,000 independently sampled donors. When a conclusion concerns differences between donors or experimental preparations, inspect how the analysis handles cells nested within those samples.
Record both the number of independent samples and the cell counts. They describe different aspects of the evidence and should not compete for one headline sample-size number.
A clear summary specifies the cell-level measurement, distribution or subgroup change, quality criteria and independent sample structure. It can then explain what the study learned about heterogeneity without converting every dot into an independent confirmation.
Sources and further detail
- Brennecke and colleagues — Accounting for technical noise in single-cell RNA-seq experiments (opens in a new tab)
Original author-hosted PDF read for technical noise and low-input variability. Publisher correction also checked: it concerns an ERCC dilution, which is not reproduced here. No operating protocol provided.
- Squair and colleagues — Confronting false discoveries in single-cell differential expression (opens in a new tab)
Original 2021 study results and discussion read for biological-replicate variation. No universal software ranking claimed. All distribution, mixture and cell-count examples are original.
Sources checked 19 September 2026. Worked examples are illustrative unless a supplied report is explicitly identified. This article has not undergone independent scientific peer review.