eduKateSG · Why Science?
Keep donors visible while asking whether a cell population changes state across conditions—because thousands of cells do not replace biological replication
Reading routes
Science learning becomes useful when a familiar object or observation is turned into a system of quantities, mechanisms and claim limits. This guide owns one applied evidence-reading job inside eduKateSG’s wider Science estate. It connects naturally to Why Science Single Cell Rna Sequencing Barcodes Transcriptome Heterogeneity Evidence; Why Science Diffcyt High Resolution Clustering Differential Discovery; Why Science Masc Mixed Effects Disease Associated Cell Populations; Why Science Sccomp Differential Composition Variability Single Cell Data; Education Hub; Singapore Secondary School Directory; Career Adulthood Hub; Science Learning Hub. It also keeps current school and public claims traceable to visible primary sources: muscat primary study; muscat PubMed record; muscat open full text; Official Bioconductor muscat package; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science syllabus; 2026 MOE G2 Computing syllabus. The sources describe the scientific scope; this article translates that scope into a calm route for Primary Science, PSLE Science, Secondary Science, O-Level Science, STEM exploration, school choices and career pathways without inventing admission or employment outcomes.
muscat, described by Crowell and colleagues in Nature Communications in 2020, is a framework for multi-sample, multi-group single-cell RNA-sequencing analysis. It supports differential-state testing within cell subpopulations using pseudobulk and cell-level mixed-model approaches, and it includes simulation tools for benchmarking complex experimental designs. Its central lesson is wonderfully practical: cells are measurements nested inside biological samples, so the experimental unit, design matrix and sample balance must remain visible from question to claim.
Inside this guide
1–12 · Foundations and models
- 1. Begin with samples, groups and subpopulations
- 2. Define differential state
- 3. Protect the experimental unit
- 4. Name the comparison
- 5. Separate abundance from state
- 6. Plan multi-factor designs carefully
- 7. Read the 2020 evidence in scope
- 8. Build the sample metadata first
- 9. Quality-control without reading outcomes
- 10. Define clusters reproducibly
- 11. Check every cluster-by-sample combination
- 12. Choose gene filters visibly
13–24 · Evidence, testing and applications
- 13. Aggregate pseudobulk counts
- 14. Build the design matrix
- 15. Practise with a fictional muscat table
- 16. Choose a pseudobulk engine
- 17. Understand mixed-model alternatives
- 18. Adjust across the declared family
- 19. Read effect size before rank
- 20. Inspect sample-level agreement
- 21. Test cluster stability
- 22. Check library depth and composition
- 23. Compare pseudobulk and mixed models
- 24. Use simulation as a question, not decoration
25–36 · Learning, decisions and pathways
- 25. Compare with diffcyt
- 26. Compare with MAST
- 27. Build a differential-state evidence card
- 28. Design an independent validation
- 29. Connect muscat to school science
- 30. Try a nested-samples activity
- 31. Connect to mathematics
- 32. Connect to computing and AI literacy
- 33. Connect to school and career pathways
- 34. Create a family evidence habit
- 35. Use precise muscat language
- 36. Finish with a sample-to-claim bundle
Section 1 of 36
1. Begin with samples, groups and subpopulations
A single-cell experiment may compare treatments across several people while measuring many cell types. The useful question is not merely which genes vary, but where, between which groups and across which independent samples.
muscat preserves the multi-sample design by summarising or modelling cells within cluster-by-sample combinations before comparing experimental groups.
Archive sample manifests, preprocessing decisions, model formulae, contrasts, random seeds, package versions and exclusions so another analyst can reconstruct the decision path.
Section 2 of 36
2. Define differential state
A differential-state gene changes expression within a selected cell subpopulation across experimental conditions. This separates cellular state from the number of cells assigned to that population.
Differential state asks whether expression changes within a cell subpopulation; it is distinct from differential abundance, which asks whether that subpopulation changes in representation.
Keep the biological sample as the unit of replication. More cells can sharpen cellular description, but they do not manufacture more independent donors, animals or cultures.
Section 3 of 36
3. Protect the experimental unit
If ten thousand cells came from one donor, the experiment still has one donor for population-level inference. Cells carry resolution; donors, animals or cultures carry replication.
Pseudobulk workflows aggregate counts for each gene, sample and cluster, then use established count-based models with the biological sample as the replicate.
Use negative controls, simulated nulls or label permutations only when their exchangeability assumptions match the design, and state exactly what false signal each check can reveal.
Section 4 of 36
4. Name the comparison
Write the exact group contrast, cell population, tissue and time point before modelling. A vague search across every possible comparison invites post-hoc storytelling.
Cell-level mixed models can represent within-sample structure but are computationally heavier and still require enough independent samples to support group inference.
Report effect direction, uncertainty, sample support and sensitivity beside adjusted evidence; a ranked gene list or colourful curve is not a complete result.
Section 5 of 36
5. Separate abundance from state
A cluster can become more common without changing its gene expression, or change expression without becoming more common. Analyse and describe those outcomes separately.
The muscat paper benchmarked methods using simulated multi-sample data, showing that performance depends on effect type, cluster frequency, sample size and model choice.
Create an evidence card naming the comparison, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 6 of 36
6. Plan multi-factor designs carefully
Batch, donor pairing, treatment, sex, age or time may matter, but the sample count limits how many terms can be estimated responsibly.
A cluster label is an analysis object, not automatically a stable cell type. Marker coherence, sample coverage and sensitivity to clustering should accompany interpretation.
Write association, distributional difference or trajectory-linked pattern when that is what was tested. Reserve cause, diagnosis and benefit for designs that support those stronger claims.
Section 7 of 36
7. Read the 2020 evidence in scope
Crowell and colleagues presented muscat for differential-state analysis in multi-sample, multi-condition single-cell data and compared pseudobulk with cell-level mixed-model strategies through benchmark simulations and applications.
muscat preserves the multi-sample design by summarising or modelling cells within cluster-by-sample combinations before comparing experimental groups.
Archive sample manifests, preprocessing decisions, model formulae, contrasts, random seeds, package versions and exclusions so another analyst can reconstruct the decision path.
Section 8 of 36
8. Build the sample metadata first
Create one row per biological sample with group, batch, pairing, tissue, collection, processing and exclusion fields. Validate that every cell barcode maps to exactly one sample.
Differential state asks whether expression changes within a cell subpopulation; it is distinct from differential abundance, which asks whether that subpopulation changes in representation.
Keep the biological sample as the unit of replication. More cells can sharpen cellular description, but they do not manufacture more independent donors, animals or cultures.
Section 9 of 36
9. Quality-control without reading outcomes
Apply cell-quality, doublet, ambient-RNA and low-count rules consistently. Outcome-aware filtering can manufacture a cleaner difference than the experiment produced.
Pseudobulk workflows aggregate counts for each gene, sample and cluster, then use established count-based models with the biological sample as the replicate.
Use negative controls, simulated nulls or label permutations only when their exchangeability assumptions match the design, and state exactly what false signal each check can reveal.
Section 10 of 36
10. Define clusters reproducibly
Record dimensionality reduction, neighbour graph, resolution, markers and annotation rules. Repeat key results across defensible nearby clustering choices.
Cell-level mixed models can represent within-sample structure but are computationally heavier and still require enough independent samples to support group inference.
Report effect direction, uncertainty, sample support and sensitivity beside adjusted evidence; a ranked gene list or colourful curve is not a complete result.
Section 11 of 36
11. Check every cluster-by-sample combination
A cluster observed in one group but nearly absent from another may not support stable state testing. Plot cells and summed counts per sample before fitting models.
The muscat paper benchmarked methods using simulated multi-sample data, showing that performance depends on effect type, cluster frequency, sample size and model choice.
Create an evidence card naming the comparison, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 12 of 36
12. Choose gene filters visibly
Remove genes with insufficient counts using a declared rule applied without chasing significance. State whether filtering occurs globally or within clusters.
A cluster label is an analysis object, not automatically a stable cell type. Marker coherence, sample coverage and sensitivity to clustering should accompany interpretation.
Write association, distributional difference or trajectory-linked pattern when that is what was tested. Reserve cause, diagnosis and benefit for designs that support those stronger claims.
Section 13 of 36
13. Aggregate pseudobulk counts
For each sample and cluster, sum raw counts gene by gene. This produces a familiar sample-level count matrix for each subpopulation.
muscat preserves the multi-sample design by summarising or modelling cells within cluster-by-sample combinations before comparing experimental groups.
Archive sample manifests, preprocessing decisions, model formulae, contrasts, random seeds, package versions and exclusions so another analyst can reconstruct the decision path.
Section 14 of 36
14. Build the design matrix
Represent group, pairing and justified covariates at sample level. Ensure the coefficient or contrast matches the written biological question.
Differential state asks whether expression changes within a cell subpopulation; it is distinct from differential abundance, which asks whether that subpopulation changes in representation.
Keep the biological sample as the unit of replication. More cells can sharpen cellular description, but they do not manufacture more independent donors, animals or cultures.
Section 15 of 36
15. Practise with a fictional muscat table
This classroom table is invented and is not a result from the muscat paper.
| Fictional cluster | Gene | Log effect | Adjusted evidence | Samples represented |
|---|---|---|---|---|
| Memory T | GENE-A | 1.10 | 0.012 | 12 |
| Monocyte | GENE-B | -0.63 | 0.071 | 11 |
| Rare C | GENE-C | 2.40 | 0.030 | 3 |
Pseudobulk workflows aggregate counts for each gene, sample and cluster, then use established count-based models with the biological sample as the replicate.
Use negative controls, simulated nulls or label permutations only when their exchangeability assumptions match the design, and state exactly what false signal each check can reveal.
Section 16 of 36
16. Choose a pseudobulk engine
Count-based tools can model aggregated cluster-sample matrices with library-size handling and dispersion estimation. The exact workflow and contrast should be reported, not hidden behind a default call.
Cell-level mixed models can represent within-sample structure but are computationally heavier and still require enough independent samples to support group inference.
Report effect direction, uncertainty, sample support and sensitivity beside adjusted evidence; a ranked gene list or colourful curve is not a complete result.
Section 17 of 36
17. Understand mixed-model alternatives
Cell-level models can include sample effects while retaining cells as observations. Inspect convergence, computational stability and assumptions rather than treating added complexity as automatic superiority.
The muscat paper benchmarked methods using simulated multi-sample data, showing that performance depends on effect type, cluster frequency, sample size and model choice.
Create an evidence card naming the comparison, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 18 of 36
18. Adjust across the declared family
Thousands of genes may be tested in several clusters. Define whether correction occurs within a cluster, across all cluster-gene tests or in another predeclared family.
A cluster label is an analysis object, not automatically a stable cell type. Marker coherence, sample coverage and sensitivity to clustering should accompany interpretation.
Write association, distributional difference or trajectory-linked pattern when that is what was tested. Reserve cause, diagnosis and benefit for designs that support those stronger claims.
Section 19 of 36
19. Read effect size before rank
A small adjusted value can accompany a tiny expression shift, while a large shift in a rare cluster may be uncertain. Show estimates and intervals with evidence values.
muscat preserves the multi-sample design by summarising or modelling cells within cluster-by-sample combinations before comparing experimental groups.
Archive sample manifests, preprocessing decisions, model formulae, contrasts, random seeds, package versions and exclusions so another analyst can reconstruct the decision path.
Section 20 of 36
20. Inspect sample-level agreement
Plot pseudobulk values or residual summaries for every donor. A pooled-looking difference may be carried by one sample or a batch-specific subgroup.
Differential state asks whether expression changes within a cell subpopulation; it is distinct from differential abundance, which asks whether that subpopulation changes in representation.
Keep the biological sample as the unit of replication. More cells can sharpen cellular description, but they do not manufacture more independent donors, animals or cultures.
Section 21 of 36
21. Test cluster stability
Repeat the analysis after modest changes to resolution or annotation. A credible signal should map to a marker-coherent biological region rather than one arbitrary cluster number.
Pseudobulk workflows aggregate counts for each gene, sample and cluster, then use established count-based models with the biological sample as the replicate.
Use negative controls, simulated nulls or label permutations only when their exchangeability assumptions match the design, and state exactly what false signal each check can reveal.
Section 22 of 36
22. Check library depth and composition
Cluster-sample totals can differ greatly. Review offsets, normalisation assumptions and whether a few highly expressed genes dominate the count budget.
Cell-level mixed models can represent within-sample structure but are computationally heavier and still require enough independent samples to support group inference.
Report effect direction, uncertainty, sample support and sensitivity beside adjusted evidence; a ranked gene list or colourful curve is not a complete result.
Section 23 of 36
23. Compare pseudobulk and mixed models
Agreement across reasonable engines increases confidence; disagreement is diagnostic. Examine sparsity, variance assumptions, sample imbalance and random-effect estimation before choosing a story.
The muscat paper benchmarked methods using simulated multi-sample data, showing that performance depends on effect type, cluster frequency, sample size and model choice.
Create an evidence card naming the comparison, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 24 of 36
24. Use simulation as a question, not decoration
Simulate designs resembling the planned sample counts, cell frequencies and effect sizes to estimate what the study can detect and where false discoveries may arise.
A cluster label is an analysis object, not automatically a stable cell type. Marker coherence, sample coverage and sensitivity to clustering should accompany interpretation.
Write association, distributional difference or trajectory-linked pattern when that is what was tested. Reserve cause, diagnosis and benefit for designs that support those stronger claims.
Section 25 of 36
25. Compare with diffcyt
diffcyt addresses differential abundance and state in cytometry through high-resolution clusters. muscat focuses on multi-sample single-cell transcriptomic designs and gene-level state tests.
muscat preserves the multi-sample design by summarising or modelling cells within cluster-by-sample combinations before comparing experimental groups.
Archive sample manifests, preprocessing decisions, model formulae, contrasts, random seeds, package versions and exclusions so another analyst can reconstruct the decision path.
Section 26 of 36
26. Compare with MAST
MAST models detection and positive expression at cell level. muscat provides an experiment-aware framework for multi-sample, multi-group comparisons and can place cell-level approaches beside pseudobulk alternatives.
Differential state asks whether expression changes within a cell subpopulation; it is distinct from differential abundance, which asks whether that subpopulation changes in representation.
Keep the biological sample as the unit of replication. More cells can sharpen cellular description, but they do not manufacture more independent donors, animals or cultures.
Section 27 of 36
27. Build a differential-state evidence card
Record cluster markers, sample counts, group contrast, aggregation or mixed-model choice, effect, uncertainty, multiplicity family, sensitivity and replication status.
Pseudobulk workflows aggregate counts for each gene, sample and cluster, then use established count-based models with the biological sample as the replicate.
Use negative controls, simulated nulls or label permutations only when their exchangeability assumptions match the design, and state exactly what false signal each check can reveal.
Section 28 of 36
28. Design an independent validation
Test a short gene set in new biological samples with targeted RNA measurement, protein assays or functional experiments that match the proposed interpretation.
Cell-level mixed models can represent within-sample structure but are computationally heavier and still require enough independent samples to support group inference.
Report effect direction, uncertainty, sample support and sensitivity beside adjusted evidence; a ranked gene list or colourful curve is not a complete result.
Section 29 of 36
29. Connect muscat to school science
The method makes fair testing concrete: many readings from one specimen improve detail, while repeated independent specimens support a general comparison.
The muscat paper benchmarked methods using simulated multi-sample data, showing that performance depends on effect type, cluster frequency, sample size and model choice.
Create an evidence card naming the comparison, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 30 of 36
30. Try a nested-samples activity
Measure several leaves from each plant in treated and untreated groups. Compare a leaf-naive analysis with plant-level summaries and discuss why the confidence changes.
A cluster label is an analysis object, not automatically a stable cell type. Marker coherence, sample coverage and sensitivity to clustering should accompany interpretation.
Write association, distributional difference or trajectory-linked pattern when that is what was tested. Reserve cause, diagnosis and benefit for designs that support those stronger claims.
Section 31 of 36
31. Connect to mathematics
Nested data, sums, offsets, contrasts, variance and false-discovery control show how a complex table becomes a bounded conclusion.
muscat preserves the multi-sample design by summarising or modelling cells within cluster-by-sample combinations before comparing experimental groups.
Archive sample manifests, preprocessing decisions, model formulae, contrasts, random seeds, package versions and exclusions so another analyst can reconstruct the decision path.
Section 32 of 36
32. Connect to computing and AI literacy
A pipeline can process millions of values, yet every label, filter, cluster and design term encodes a human decision that must be documented.
Differential state asks whether expression changes within a cell subpopulation; it is distinct from differential abundance, which asks whether that subpopulation changes in representation.
Keep the biological sample as the unit of replication. More cells can sharpen cellular description, but they do not manufacture more independent donors, animals or cultures.
Section 33 of 36
33. Connect to school and career pathways
Interest in cells and statistics can lead toward biology, biotechnology, medicine, public health, computing or bioinformatics through different education routes.
Pseudobulk workflows aggregate counts for each gene, sample and cluster, then use established count-based models with the biological sample as the replicate.
Use negative controls, simulated nulls or label permutations only when their exchangeability assumptions match the design, and state exactly what false signal each check can reveal.
Section 34 of 36
34. Create a family evidence habit
When a headline cites thousands of cells, ask how many independent people or specimens contributed them and whether most samples showed the effect.
Cell-level mixed models can represent within-sample structure but are computationally heavier and still require enough independent samples to support group inference.
Report effect direction, uncertainty, sample support and sensitivity beside adjusted evidence; a ranked gene list or colourful curve is not a complete result.
Section 35 of 36
35. Use precise muscat language
Say differential state within an annotated subpopulation under the fitted design. Avoid calling a gene causal, diagnostic or universal without separate evidence.
The muscat paper benchmarked methods using simulated multi-sample data, showing that performance depends on effect type, cluster frequency, sample size and model choice.
Create an evidence card naming the comparison, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 36 of 36
36. Finish with a sample-to-claim bundle
Deliver manifests, quality checks, cluster definitions, count matrices, design and contrasts, model diagnostics, sample plots, multiplicity control, sensitivity analyses and validation plans.
A cluster label is an analysis object, not automatically a stable cell type. Marker coherence, sample coverage and sensitivity to clustering should accompany interpretation.
Write association, distributional difference or trajectory-linked pattern when that is what was tested. Reserve cause, diagnosis and benefit for designs that support those stronger claims.
## Source date and scope note Primary papers, open records and official software documentation were checked for this article on 10 October 2026. Software interfaces and recommendations can change, so readers should consult the linked current documentation before reproducing an analysis.
## Final reader checklist Before accepting a claim, confirm the biological sample count, condition definition, measured modality, preprocessing, statistical unit, model formula, uncertainty, software version, negative controls, sensitivity checks, independent validation and the exact boundary between association and causation.
Contents · Previous section · Continue to the Science Learning Hub