eduKateSG · Why Science?
Test which relative cell-type abundances change when every proportion must share the same whole—and make the reference explicit
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 Milo Neighbourhood Differential Abundance Single Cell Population Shifts; Why Science Augur Cell Type Responsiveness Perturbation Priority Evidence; Why Science Mixscape Perturbation Detection Crispr Response Signatures; Why Science Cinema Ot Confounder Separation Single Cell Treatment Effects; 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: scCODA primary study; scCODA PubMed record; Official scCODA repository; 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.
scCODA, presented by Büttner and colleagues in Nature Communications in 2021, is a Bayesian framework for analysing cell-type count compositions. It models sample-level counts while respecting the fact that cell-type proportions sum to one, uses a reference component for relative interpretation and applies model selection designed to identify credible changes while controlling false discoveries. This is especially useful when biological replicates are limited, but it does not create information that the study did not measure. Results are relative to the reference, sampling process and model assumptions.
Inside this guide
1–12 · Foundations and models
- 1. Begin with a fixed whole
- 2. Use counts, not isolated percentages
- 3. Choose a reference component
- 4. Model the composition jointly
- 5. Select credible effects
- 6. Keep uncertainty visible
- 7. Read the 2021 evidence in scope
- 8. Define the biological sample
- 9. Build a count manifest
- 10. Audit cell-type annotation
- 11. Choose a reference scientifically
- 12. Test alternative references
13–24 · Evidence, testing and applications
- 13. Specify covariates and pairing
- 14. Check prior assumptions
- 15. Practise with a fictional scCODA table
- 16. Read a selected effect
- 17. Translate signs carefully
- 18. Inspect sample-level counts
- 19. Check convergence
- 20. Check predictive fit
- 21. Handle rare cell types cautiously
- 22. Run condition permutations
- 23. Separate relative and absolute abundance
- 24. Investigate mechanisms separately
25–36 · Learning, decisions and pathways
- 25. Compare with Milo
- 26. Compare with Augur
- 27. Build a composition evidence card
- 28. Design the next experiment
- 29. Connect scCODA to school science
- 30. Try a shared-pie activity
- 31. Connect to mathematics
- 32. Connect to computing and AI literacy
- 33. Connect to school and career choices
- 34. Create a family denominator habit
- 35. Use precise scCODA language
- 36. Finish with a count-to-credible-claim bundle
Section 1 of 36
1. Begin with a fixed whole
If one cell type occupies more of a sampled total, at least one other type must occupy less. Proportions cannot vary independently.
Cell-type counts form a composition because proportions within a sample sum to one; changing one component necessarily changes at least one other proportion.
Archive raw sample-by-cell-type counts, annotations, totals, covariates, reference choice, priors, chains, diagnostics, seeds and software version.
Section 2 of 36
2. Use counts, not isolated percentages
Retain the sample-by-cell-type count table and total cells. Percentages alone hide sampling depth and the discrete evidence that generated them.
scCODA models sample-level count vectors with a Bayesian framework rather than testing each proportion as if it were independent.
Use independent biological samples, preserve pairing and batch, and choose the reference before inspecting the most attractive effects.
Section 3 of 36
3. Choose a reference component
scCODA expresses covariate effects relative to a reference cell type intended to remain comparatively stable across conditions.
Effects are interpreted relative to a selected reference cell type, making the denominator a visible part of every biological statement.
Compare alternative references, count-based descriptive plots, label permutations and other compositional models on the same samples.
Section 4 of 36
4. Model the composition jointly
The Bayesian framework treats the vector of counts together, respecting competition for the same total rather than running unrelated tests.
Spike-and-slab-style model selection identifies credible covariate effects while targeting control of false discoveries under the model.
Report posterior inclusion, credible effects, uncertainty, convergence, predictive fit, sample-level counts and reference sensitivity.
Section 5 of 36
5. Select credible effects
Model selection separates effects supported by the posterior from those left at zero under declared priors and a false-discovery target.
Limited replicates create wide uncertainty that many cells cannot erase; sample number and design remain the evidence foundation.
Build a composition card with denominator, sample support, credible changes, diagnostics, alternatives and validation status.
Section 6 of 36
6. Keep uncertainty visible
Posterior distributions show a range of plausible effects. A concise selected set should not conceal weak sample support or poor model fit.
A relative abundance result does not distinguish proliferation, death, migration, capture bias or compositional displacement without further measurements.
Say credible relative abundance change under the scCODA model; avoid absolute expansion, depletion or causation without supporting evidence.
Section 7 of 36
7. Read the 2021 evidence in scope
Büttner and colleagues introduced scCODA for compositional single-cell data analysis, with evaluations across simulated and biological datasets.
Cell-type counts form a composition because proportions within a sample sum to one; changing one component necessarily changes at least one other proportion.
Archive raw sample-by-cell-type counts, annotations, totals, covariates, reference choice, priors, chains, diagnostics, seeds and software version.
Section 8 of 36
8. Define the biological sample
The replicate is a donor, animal, culture or specimen. Individual cells within one sample share conditions and are not independent replicates.
scCODA models sample-level count vectors with a Bayesian framework rather than testing each proportion as if it were independent.
Use independent biological samples, preserve pairing and batch, and choose the reference before inspecting the most attractive effects.
Section 9 of 36
9. Build a count manifest
Record sample, condition, batch, tissue, cell-type labels, total captured cells, exclusions and any pairing before calculating proportions.
Effects are interpreted relative to a selected reference cell type, making the denominator a visible part of every biological statement.
Compare alternative references, count-based descriptive plots, label permutations and other compositional models on the same samples.
Section 10 of 36
10. Audit cell-type annotation
Marker choices, reference mapping and resolution determine the columns of the composition. Merge uncertain rare categories before making precise claims.
Spike-and-slab-style model selection identifies credible covariate effects while targeting control of false discoveries under the model.
Report posterior inclusion, credible effects, uncertainty, convergence, predictive fit, sample-level counts and reference sensitivity.
Section 11 of 36
11. Choose a reference scientifically
Prefer a sufficiently abundant component expected to be stable, then justify that expectation independently of the desired result.
Limited replicates create wide uncertainty that many cells cannot erase; sample number and design remain the evidence foundation.
Build a composition card with denominator, sample support, credible changes, diagnostics, alternatives and validation status.
Section 12 of 36
12. Test alternative references
Because effects are relative, rerun defensible reference choices and translate coefficients carefully. Robust biological conclusions should not hinge on one denominator.
A relative abundance result does not distinguish proliferation, death, migration, capture bias or compositional displacement without further measurements.
Say credible relative abundance change under the scCODA model; avoid absolute expansion, depletion or causation without supporting evidence.
Section 13 of 36
13. Specify covariates and pairing
Include condition, batch and identifiable design factors. Preserve matched samples where the experiment provides them.
Cell-type counts form a composition because proportions within a sample sum to one; changing one component necessarily changes at least one other proportion.
Archive raw sample-by-cell-type counts, annotations, totals, covariates, reference choice, priors, chains, diagnostics, seeds and software version.
Section 14 of 36
14. Check prior assumptions
Priors influence shrinkage and effect selection, especially with few samples. Report defaults, changes and sensitivity rather than treating them as invisible.
scCODA models sample-level count vectors with a Bayesian framework rather than testing each proportion as if it were independent.
Use independent biological samples, preserve pairing and batch, and choose the reference before inspecting the most attractive effects.
Section 15 of 36
15. Practise with a fictional scCODA table
This classroom table is invented and is not output from the scCODA paper.
| Fictional cell type | Relative effect | Credible | Reference sensitivity | Next action |
|---|---|---|---|---|
| Type A | 0.8 | Yes | Stable | Validate counts |
| Type B | -0.4 | No | Variable | Add samples |
| Type C | 0.0 | Reference | Not applicable | Monitor |
Effects are interpreted relative to a selected reference cell type, making the denominator a visible part of every biological statement.
Compare alternative references, count-based descriptive plots, label permutations and other compositional models on the same samples.
Section 16 of 36
16. Read a selected effect
A credible non-zero coefficient indicates a condition-associated change relative to the reference under the model and coding used.
Spike-and-slab-style model selection identifies credible covariate effects while targeting control of false discoveries under the model.
Report posterior inclusion, credible effects, uncertainty, convergence, predictive fit, sample-level counts and reference sensitivity.
Section 17 of 36
17. Translate signs carefully
The sign depends on covariate and reference levels. State which condition is compared with which and avoid ambiguous increase language.
Limited replicates create wide uncertainty that many cells cannot erase; sample number and design remain the evidence foundation.
Build a composition card with denominator, sample support, credible changes, diagnostics, alternatives and validation status.
Section 18 of 36
18. Inspect sample-level counts
Plot every sample. A posterior summary without raw count patterns can conceal an outlier or batch-confounded separation.
A relative abundance result does not distinguish proliferation, death, migration, capture bias or compositional displacement without further measurements.
Say credible relative abundance change under the scCODA model; avoid absolute expansion, depletion or causation without supporting evidence.
Section 19 of 36
19. Check convergence
Review chains and diagnostic summaries where available. A model that has not sampled its posterior reliably cannot support a polished biological story.
Cell-type counts form a composition because proportions within a sample sum to one; changing one component necessarily changes at least one other proportion.
Archive raw sample-by-cell-type counts, annotations, totals, covariates, reference choice, priors, chains, diagnostics, seeds and software version.
Section 20 of 36
20. Check predictive fit
Posterior predictive checks ask whether simulated counts resemble important features of observed compositions, including variation and zeros.
scCODA models sample-level count vectors with a Bayesian framework rather than testing each proportion as if it were independent.
Use independent biological samples, preserve pairing and batch, and choose the reference before inspecting the most attractive effects.
Section 21 of 36
21. Handle rare cell types cautiously
Sparse categories can yield unstable relative effects. Consider justified aggregation and report detection limitations.
Effects are interpreted relative to a selected reference cell type, making the denominator a visible part of every biological statement.
Compare alternative references, count-based descriptive plots, label permutations and other compositional models on the same samples.
Section 22 of 36
22. Run condition permutations
Shuffle labels at the sample level within valid blocks. The full pipeline should not routinely select effects under the null.
Spike-and-slab-style model selection identifies credible covariate effects while targeting control of false discoveries under the model.
Report posterior inclusion, credible effects, uncertainty, convergence, predictive fit, sample-level counts and reference sensitivity.
Section 23 of 36
23. Separate relative and absolute abundance
A rise in proportion can occur even if the cell count stays constant while another population falls. Absolute quantification needs additional measurement.
Limited replicates create wide uncertainty that many cells cannot erase; sample number and design remain the evidence foundation.
Build a composition card with denominator, sample support, credible changes, diagnostics, alternatives and validation status.
Section 24 of 36
24. Investigate mechanisms separately
Proliferation, death, migration, differentiation and capture bias can produce similar compositional patterns. Design follow-up assays that discriminate them.
A relative abundance result does not distinguish proliferation, death, migration, capture bias or compositional displacement without further measurements.
Say credible relative abundance change under the scCODA model; avoid absolute expansion, depletion or causation without supporting evidence.
Section 25 of 36
25. Compare with Milo
Milo tests local graph-neighbourhood counts; scCODA tests predeclared cell-type count compositions with an explicit reference component.
Cell-type counts form a composition because proportions within a sample sum to one; changing one component necessarily changes at least one other proportion.
Archive raw sample-by-cell-type counts, annotations, totals, covariates, reference choice, priors, chains, diagnostics, seeds and software version.
Section 26 of 36
26. Compare with Augur
Augur asks how separable perturbation conditions are within cell types. scCODA asks which relative cell-type abundances credibly change across samples.
scCODA models sample-level count vectors with a Bayesian framework rather than testing each proportion as if it were independent.
Use independent biological samples, preserve pairing and batch, and choose the reference before inspecting the most attractive effects.
Section 27 of 36
27. Build a composition evidence card
Place raw counts, reference rationale, model formula, selected effects, diagnostics, alternative references and absolute-measurement status together.
Effects are interpreted relative to a selected reference cell type, making the denominator a visible part of every biological statement.
Compare alternative references, count-based descriptive plots, label permutations and other compositional models on the same samples.
Section 28 of 36
28. Design the next experiment
Add independent samples and orthogonal absolute counts, standardise recovery and predeclare the reference before seeing outcomes.
Spike-and-slab-style model selection identifies credible covariate effects while targeting control of false discoveries under the model.
Report posterior inclusion, credible effects, uncertainty, convergence, predictive fit, sample-level counts and reference sensitivity.
Section 29 of 36
29. Connect scCODA to school science
Fractions, ecosystems and limiting totals provide an intuitive bridge to compositional reasoning in modern biology.
Limited replicates create wide uncertainty that many cells cannot erase; sample number and design remain the evidence foundation.
Build a composition card with denominator, sample support, credible changes, diagnostics, alternatives and validation status.
Section 30 of 36
30. Try a shared-pie activity
Give groups fixed totals of coloured counters, alter one category and observe how every percentage changes even when some counts do not.
A relative abundance result does not distinguish proliferation, death, migration, capture bias or compositional displacement without further measurements.
Say credible relative abundance change under the scCODA model; avoid absolute expansion, depletion or causation without supporting evidence.
Section 31 of 36
31. Connect to mathematics
Ratios, logarithms, probability distributions, Bayesian updating and false-discovery control make relative evidence explicit.
Cell-type counts form a composition because proportions within a sample sum to one; changing one component necessarily changes at least one other proportion.
Archive raw sample-by-cell-type counts, annotations, totals, covariates, reference choice, priors, chains, diagnostics, seeds and software version.
Section 32 of 36
32. Connect to computing and AI literacy
A sophisticated model cannot infer absolute change from relative counts alone. The data structure constrains the answer.
scCODA models sample-level count vectors with a Bayesian framework rather than testing each proportion as if it were independent.
Use independent biological samples, preserve pairing and batch, and choose the reference before inspecting the most attractive effects.
Section 33 of 36
33. Connect to school and career choices
Learners who enjoy parts and wholes can explore biology, statistics, computing, public health or bioinformatics through many education paths.
Effects are interpreted relative to a selected reference cell type, making the denominator a visible part of every biological statement.
Compare alternative references, count-based descriptive plots, label permutations and other compositional models on the same samples.
Section 34 of 36
34. Create a family denominator habit
Whenever a percentage changes, ask what the total is, what serves as reference and whether absolute counts were measured.
Spike-and-slab-style model selection identifies credible covariate effects while targeting control of false discoveries under the model.
Report posterior inclusion, credible effects, uncertainty, convergence, predictive fit, sample-level counts and reference sensitivity.
Section 35 of 36
35. Use precise scCODA language
Say posterior-supported relative abundance effect versus the named reference. Avoid absolute growth, loss or causal mechanism without added evidence.
Limited replicates create wide uncertainty that many cells cannot erase; sample number and design remain the evidence foundation.
Build a composition card with denominator, sample support, credible changes, diagnostics, alternatives and validation status.
Section 36 of 36
36. Finish with a count-to-credible-claim bundle
Deliver count manifests, reference rationales, model diagnostics, sensitivity analyses, evidence cards and independent absolute validation.
A relative abundance result does not distinguish proliferation, death, migration, capture bias or compositional displacement without further measurements.
Say credible relative abundance change under the scCODA model; avoid absolute expansion, depletion or causation without supporting evidence.
Did you know? If a sample contains 20% type A and type A doubles in absolute number while all other counts stay fixed, every other percentage falls even though those cells did not disappear. Composition changes the language of evidence.
A useful final dashboard names the reference, displays raw counts by sample, shows posterior-selected effects and repeats the analysis with defensible alternative references.
Absolute counting beads, histology or another independent measurement can distinguish a truly changing population from a relative shift caused elsewhere in the composition.
Contents · Previous section · Continue to the Science Learning Hub