BRM5012 Chap.8 Single-cell Quality Control
Single-cell Quality Control
Single-cell quality control interprets nFeature, nCount and mitochondrial-read proportion together. Low values can indicate poor capture, high values can flag doublets, and high mitochondrial signal can indicate stress or damage, but tissue, treatment and preparation can change all three.
This chapter replaces universal cut-offs with contextual distributions, explicit mechanisms, reproducible rules and before-and-after sensitivity checks.
What this chapter covers
- 01
nFeature distributions
- 02
nCount distributions
- 03
Mitochondrial reads
- 04
Low-quality cells
- 05
Doublet risk
- 06
Biological context
- 07
Dataset-specific thresholds
- 08
Filter sensitivity
Worked example · free
Diagnose three synthetic cells
- +1Flag Cell A as a strong low-quality candidate because several signals indicate poor capture or damage.
- +1Flag Cell B as a doublet candidate because abundance and incompatible lineage markers support two cells.
- +1Retain Cell C provisionally because treatment biology may explain its mitochondrial signal.
- +1Compare each cell with sample-specific distributions and other quality metrics.
- +1Record the final rule and rerun summaries after filtering.
Key terms
- nFeature
- The number of distinct genes detected in one cell after expression counting.
- nCount
- The total detected RNA molecule count for one cell after UMI handling.
- Mitochondrial proportion
- The share of counted expression assigned to mitochondrial genes, interpreted as a context-dependent quality signal.
- Doublet
- A captured event containing more than one cell, producing a mixed expression profile.
Single-cell Quality Control FAQ
Why are there no universal filtering thresholds?
Metric distributions depend on tissue, cell size, cell quality, sample preparation, sequencing method, batch and biological treatment. Choose boundaries after inspecting this dataset and report them as operational decisions rather than biological constants.
Does high mitochondrial expression always mean damage?
No. Stress or dying cells can show high mitochondrial signal, but disease, injury or treatment can also make it biologically meaningful. Compare other metrics, marker coherence, sample distributions and study context before removal.
How can a doublet escape simple count filters?
Two low-quality cells can combine into an event whose total counts and detected genes look ordinary. Use marker incompatibility, dedicated diagnostics and neighbourhood context rather than relying on one high-count threshold.
Assessment move
Build a three-column table for nFeature, nCount and mitochondrial proportion. For each metric, list what low, typical and high values might mean, then add the biological caveat. Practise interpreting combinations rather than single numbers. Whenever you propose a filter, write the suspected mechanism, the supporting evidence and the downstream consequence.
Keep a cell-level QC table with metrics and pass or fail decisions so borderline cells can be restored. Compare retention by sample and condition, because disproportionate removal can change apparent cell composition. Replot the data after filtering and test whether important clusters or claims depend on the chosen boundary.
In written answers, conclude with remove, retain provisionally or investigate further, and name the check that would separate technical artefact from meaningful biology. Sketch paired distributions for each sample rather than memorising universal cut-offs. Mark cells with low features and counts, cells with unusually high totals, and cells with raised mitochondrial proportion. For each region, write at least two explanations.
Low information may reflect damage or a specialised low-RNA cell; high abundance may indicate a doublet or an active large cell; mitochondrial signal may reflect technical stress or genuine treatment biology. Add lineage markers and sample identity before choosing.
Construct three synthetic profiles that force different decisions: remove when independent signals converge on damage, investigate when mixed markers suggest a doublet, and retain provisionally when a coherent biological state explains the warning metric. Record the suspected mechanism beside every rule. Apply the rule to a copy of the cell table and summarise retention by specimen, condition and tentative cell type.
A large imbalance is a result to explain, not a formatting issue. Recalculate distributions and neighbourhoods after filtering, then restore borderline cells to see whether clusters or conclusions change. Preserve raw matrices and explicit pass or fail columns so every decision remains reversible. Compare sample-specific boundaries with a pooled boundary and identify when pooling hides a shifted preparation.
Practise writing a filter audit containing metric, context, mechanism, threshold, cells removed and sensitivity result. End with a decision at the right certainty: remove, retain with a qualification or collect another check. Name the evidence that would change that decision. This approach treats quality control as a biological inference problem and makes its effect on composition and downstream analysis visible.
Working through Single-cell Quality Control in BRM5012? Sia is AskSia’s AI Biology tutor — ask any BRM5012 Single-cell Quality Control question and get a clear, step-by-step explanation grounded in how BRM5012 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
