thebench.bio #1 Ecosystem for the Bio-Tech Industry
All Single-Cell Analysis questions

Question 16 of 20 Intermediate Scenario

An unexpected cluster appears in only one sample. How would you investigate its origin?

Short answer

Treat it as a possible artifact until shown otherwise. Check its QC metrics, doublet scores, ambient and stress signatures and the sample's processing record, then ask whether it is biology unique to that donor or condition, which needs replication to claim.

Why

A cluster present in one sample has several possible origins, and only one is new biology. Technical causes include a batch effect from separate processing, damaged cells with high mitochondrial fraction, doublets, ambient RNA and dissociation stress. Biological causes include a donor-specific state, such as an infection, a tumor clone, or a genuine response that only one condition shows.

The Bioconductor droplet chapter reminds analysts that called cells can still be damaged or dying, and that filtering should avoid removing genuinely metabolically active populations. So the check must look at both sides. A doublet cluster usually expresses markers of two unrelated lineages and has high UMI counts.

A stressed cluster shows immediate early genes and high mitochondrial fraction. A batch artifact often shifts every cell type in that sample, not just one cluster. A donor-specific biology cluster keeps distinct markers after these checks and may carry a genetic or clinical explanation.

Even then, one sample is one observation. The claim it supports is that this donor had the population, not that the condition creates it.

How to work it through

  1. Compare the cluster's UMI counts, gene counts and mitochondrial fraction with other clusters.
  2. Score doublets and check for co-expression of markers from unrelated lineages.
  3. Look for stress and ambient signatures, such as immediate early genes or abundant markers from other cell types.
  4. Check whether other cell types in the same sample are also shifted, which points to a sample-wide batch effect.
  5. Read the processing record for that sample, including time to processing and dissociation.
  6. If the cluster survives these checks, look for it in other samples at lower frequency and plan replication.

An example

In ten lung samples, a cluster of 600 cells appears only in donor 7. Its median UMI count is double that of other clusters, and 70 percent of its cells co-express epithelial and macrophage markers. A doublet detector flags most of them, and donor 7 was loaded at 16,000 cells against 8,000 for the others.

The cluster is removed as doublets, and the record is updated with the overloading.

Where the answer stops holding

A real rare state may also have high RNA content, so a high UMI count alone is not proof of doublets. Doublet detectors miss doublets of two cells of the same type. Without replicate donors, a donor-specific biological population cannot be generalized to the condition.

A strong answer covers

  • Start from technical explanations before biological ones
  • Check QC metrics, doublets, stress and ambient signatures
  • A sample-wide shift points to batch rather than one population
  • Read the sample's processing record
  • One sample cannot establish a condition-level finding

Common mistakes

  • Naming a new cell state before checking doublet and stress signatures
  • Integrating harder until the cluster disappears without explaining it
  • Generalizing a single donor's population to the whole condition