eduKateSG · Why Science?
Ask whether belonging to a cell population is associated with a sample phenotype—while accounting for people, batches and repeated cellular measurements
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 Cna Covarying Neighbourhoods Phenotype Associated Cell Populations; Why Science Milo Neighbourhood Differential Abundance Single Cell Population Shifts; Why Science Sccoda Compositional Models Credible Cell Type Abundance Changes; Why Science Mass Cytometry Metal Isotope Antibodies Single Cell Protein Evidence; 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: MASC primary study; MASC open full-text record; Official MASC 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.
MASC—mixed-effects association of single cells—was presented by Fonseka and colleagues in Science Translational Medicine in 2018. It tests whether a sample-level covariate is associated with membership of individual cells in a defined cellular subset, using mixed-effects logistic regression to account for technical covariates and biological variation. The method helped identify an expanded effector CD4 T-cell population in rheumatoid arthritis cohorts, but an association from a particular study is not automatically a diagnostic marker, a causal population or a treatment target.
Inside this guide
1–12 · Foundations and models
- 1. Begin with clustered cells and people
- 2. Ask the reverse-association question
- 3. Define the subset
- 4. Name the predictor of interest
- 5. Recognise repeated cells
- 6. Use a mixed-effects model
- 7. Read the 2018 evidence in scope
- 8. Build a participant manifest
- 9. Separate discovery cohorts
- 10. Check case and control balance
- 11. Specify fixed effects
- 12. Specify random effects
13–24 · Evidence, testing and applications
- 13. Create the membership variable
- 14. Fit the null and full models
- 15. Practise with a fictional MASC table
- 16. Read the odds ratio
- 17. Inspect the interval
- 18. Control multiple subsets
- 19. Plot donor contributions
- 20. Run leave-one-donor-out checks
- 21. Check batch-by-phenotype alignment
- 22. Examine model convergence
- 23. Validate marker identity
- 24. Compare with CNA
25–36 · Learning, decisions and pathways
- 25. Compare with scCODA
- 26. Build an association evidence card
- 27. Design an independent replication
- 28. Separate association from clinical use
- 29. Connect MASC to school science
- 30. Try a classroom random-effect 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 MASC language
- 36. Finish with a cohort-aware bundle
Section 1 of 36
1. Begin with clustered cells and people
A study may contain hundreds of thousands of cells but only dozens of participants. The cell count is visually large while the number of independent biological units remains modest.
MASC reverses the familiar viewpoint: the binary response is whether a cell belongs to a chosen subset, while the predictor of interest is a sample-level phenotype.
Archive sample manifests, preprocessing, feature roles, model formulae, random seeds, software versions and exclusions so another analyst can reconstruct the decision path.
Section 2 of 36
2. Ask the reverse-association question
For one cell at a time, ask whether membership in a candidate subset is associated with disease or another sample phenotype after accounting for design variables.
A donor or sample random intercept helps account for repeated cells from the same biological unit; it does not rescue a study with too few or perfectly confounded donors.
Keep the biological sample as the unit of replication. Many cells improve resolution, but they do not turn one donor, animal or culture into many independent experiments.
Section 3 of 36
3. Define the subset
Use a reproducible cluster, gate or marker rule. Describe whether the population was pre-specified, discovered in the same cohort or transferred from another dataset.
Fixed effects can adjust measured technical or biological covariates, but adjustment is credible only when the model is specified from subject knowledge rather than result hunting.
Use negative controls, label permutations or design-matched null analyses where valid. A trustworthy signal should exceed artefacts that the same workflow can manufacture.
Section 4 of 36
4. Name the predictor of interest
Disease status, treatment, response or continuous clinical score should be measured at a clear time and level. Avoid mixing baseline and post-treatment labels.
The cellular subset must be defined before the association is interpreted. Cluster discovery and association testing on the same data can require extra validation against selection bias.
Report effect direction, uncertainty, sample support and sensitivity—not only a colourful embedding, selected feature or adjusted value.
Section 5 of 36
5. Recognise repeated cells
Cells from one donor share genetics, environment, collection and processing. Treating them as unrelated observations yields confidence that is much too narrow.
An odds ratio describes association with subset membership under the fitted model. It is not the same as absolute prevalence, risk, diagnostic accuracy or causal effect.
Create an evidence card naming the contrast, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 6 of 36
6. Use a mixed-effects model
A donor or sample random effect captures some within-person dependence, while fixed effects represent the phenotype and declared covariates.
The rheumatoid-arthritis application motivates MASC, but every new disease, tissue and assay needs its own sampling, quality control and replication evidence.
Write association, stratification, differential abundance or differential state when that is what was tested. Reserve cause, diagnosis and benefit for evidence that directly supports them.
Section 7 of 36
7. Read the 2018 evidence in scope
Fonseka and colleagues introduced MASC and applied it across single-cell transcriptomic and mass-cytometry data to identify a disease-associated effector CD4 T-cell population in rheumatoid arthritis.
MASC reverses the familiar viewpoint: the binary response is whether a cell belongs to a chosen subset, while the predictor of interest is a sample-level phenotype.
Archive sample manifests, preprocessing, feature roles, model formulae, random seeds, software versions and exclusions so another analyst can reconstruct the decision path.
Section 8 of 36
8. Build a participant manifest
Record participant, cohort, diagnosis, treatment, tissue, collection date, batch, demographic variables, cell yield and exclusion before modelling.
A donor or sample random intercept helps account for repeated cells from the same biological unit; it does not rescue a study with too few or perfectly confounded donors.
Keep the biological sample as the unit of replication. Many cells improve resolution, but they do not turn one donor, animal or culture into many independent experiments.
Section 9 of 36
9. Separate discovery cohorts
If a population is found in one cohort, evaluate its marker definition and association in another cohort or modality where possible.
Fixed effects can adjust measured technical or biological covariates, but adjustment is credible only when the model is specified from subject knowledge rather than result hunting.
Use negative controls, label permutations or design-matched null analyses where valid. A trustworthy signal should exceed artefacts that the same workflow can manufacture.
Section 10 of 36
10. Check case and control balance
Unequal groups, recruitment sites or medications can dominate the estimate. Show counts at the participant level, not merely cells per group.
The cellular subset must be defined before the association is interpreted. Cluster discovery and association testing on the same data can require extra validation against selection bias.
Report effect direction, uncertainty, sample support and sensitivity—not only a colourful embedding, selected feature or adjusted value.
Section 11 of 36
11. Specify fixed effects
Include defensible technical and biological covariates such as batch or age when supported by design and sample size. Avoid a crowded formula that the cohort cannot estimate.
An odds ratio describes association with subset membership under the fitted model. It is not the same as absolute prevalence, risk, diagnostic accuracy or causal effect.
Create an evidence card naming the contrast, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 12 of 36
12. Specify random effects
A participant intercept is common when many cells belong to each person. Repeated visits or nested batches may require a richer structure and adequate replication at each level.
The rheumatoid-arthritis application motivates MASC, but every new disease, tissue and assay needs its own sampling, quality control and replication evidence.
Write association, stratification, differential abundance or differential state when that is what was tested. Reserve cause, diagnosis and benefit for evidence that directly supports them.
Section 13 of 36
13. Create the membership variable
Code whether each analysed cell belongs to the target subset. Exclude ambiguous or low-quality events by predeclared rules rather than outcome-aware editing.
MASC reverses the familiar viewpoint: the binary response is whether a cell belongs to a chosen subset, while the predictor of interest is a sample-level phenotype.
Archive sample manifests, preprocessing, feature roles, model formulae, random seeds, software versions and exclusions so another analyst can reconstruct the decision path.
Section 14 of 36
14. Fit the null and full models
Compare a model without the phenotype term to one that includes it, while keeping other specified effects aligned. The comparison targets the incremental association of interest.
A donor or sample random intercept helps account for repeated cells from the same biological unit; it does not rescue a study with too few or perfectly confounded donors.
Keep the biological sample as the unit of replication. Many cells improve resolution, but they do not turn one donor, animal or culture into many independent experiments.
Section 15 of 36
15. Practise with a fictional MASC table
This teaching example is invented and is not a result from the MASC paper.
| Fictional subset | Adjusted odds ratio | Interval | Donors represented | Claim |
|---|---|---|---|---|
| S4 | 1.80 | 1.25–2.58 | 22 | Associated |
| S9 | 0.72 | 0.48–1.09 | 20 | Uncertain |
| S12 | 3.10 | 1.02–9.44 | 4 | Fragile |
Fixed effects can adjust measured technical or biological covariates, but adjustment is credible only when the model is specified from subject knowledge rather than result hunting.
Use negative controls, label permutations or design-matched null analyses where valid. A trustworthy signal should exceed artefacts that the same workflow can manufacture.
Section 16 of 36
16. Read the odds ratio
An odds ratio above one means cells from phenotype-positive samples have higher adjusted odds of belonging to the subset under the model. It does not give a person-level disease probability.
The cellular subset must be defined before the association is interpreted. Cluster discovery and association testing on the same data can require extra validation against selection bias.
Report effect direction, uncertainty, sample support and sensitivity—not only a colourful embedding, selected feature or adjusted value.
Section 17 of 36
17. Inspect the interval
Wide intervals signal limited participant support, sparse membership or unstable covariate adjustment. Report them beside point estimates.
An odds ratio describes association with subset membership under the fitted model. It is not the same as absolute prevalence, risk, diagnostic accuracy or causal effect.
Create an evidence card naming the contrast, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 18 of 36
18. Control multiple subsets
Testing many clusters creates multiple opportunities for chance findings. Declare the correction family and separate confirmatory subsets from exploratory searches.
The rheumatoid-arthritis application motivates MASC, but every new disease, tissue and assay needs its own sampling, quality control and replication evidence.
Write association, stratification, differential abundance or differential state when that is what was tested. Reserve cause, diagnosis and benefit for evidence that directly supports them.
Section 19 of 36
19. Plot donor contributions
Show each participant’s subset proportion or model contribution. A cell-rich outlier can otherwise look like broad cohort agreement.
MASC reverses the familiar viewpoint: the binary response is whether a cell belongs to a chosen subset, while the predictor of interest is a sample-level phenotype.
Archive sample manifests, preprocessing, feature roles, model formulae, random seeds, software versions and exclusions so another analyst can reconstruct the decision path.
Section 20 of 36
20. Run leave-one-donor-out checks
Refit after removing each participant. If direction or significance depends on one donor, describe the result as fragile and exploratory.
A donor or sample random intercept helps account for repeated cells from the same biological unit; it does not rescue a study with too few or perfectly confounded donors.
Keep the biological sample as the unit of replication. Many cells improve resolution, but they do not turn one donor, animal or culture into many independent experiments.
Section 21 of 36
21. Check batch-by-phenotype alignment
When all cases and controls were processed separately, the model cannot reliably distinguish biology from batch regardless of cell count.
Fixed effects can adjust measured technical or biological covariates, but adjustment is credible only when the model is specified from subject knowledge rather than result hunting.
Use negative controls, label permutations or design-matched null analyses where valid. A trustworthy signal should exceed artefacts that the same workflow can manufacture.
Section 22 of 36
22. Examine model convergence
Warnings, singular fits or extreme coefficients can reveal separation and insufficient information. Do not report a polished odds ratio from an unstable fit.
The cellular subset must be defined before the association is interpreted. Cluster discovery and association testing on the same data can require extra validation against selection bias.
Report effect direction, uncertainty, sample support and sensitivity—not only a colourful embedding, selected feature or adjusted value.
Section 23 of 36
23. Validate marker identity
Confirm that the associated cluster expresses a coherent marker program and maps to a reproducible population across cohorts or platforms.
An odds ratio describes association with subset membership under the fitted model. It is not the same as absolute prevalence, risk, diagnostic accuracy or causal effect.
Create an evidence card naming the contrast, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 24 of 36
24. Compare with CNA
CNA performs a global and local cohort association through covarying neighbourhood abundance. MASC tests membership in specified subsets using cell-level mixed-effects logistic regression.
The rheumatoid-arthritis application motivates MASC, but every new disease, tissue and assay needs its own sampling, quality control and replication evidence.
Write association, stratification, differential abundance or differential state when that is what was tested. Reserve cause, diagnosis and benefit for evidence that directly supports them.
Section 25 of 36
25. Compare with scCODA
scCODA models the joint composition of predefined cell categories. MASC focuses on adjusted association for one subset at a time and therefore needs explicit multiplicity handling.
MASC reverses the familiar viewpoint: the binary response is whether a cell belongs to a chosen subset, while the predictor of interest is a sample-level phenotype.
Archive sample manifests, preprocessing, feature roles, model formulae, random seeds, software versions and exclusions so another analyst can reconstruct the decision path.
Section 26 of 36
26. Build an association evidence card
Record subset definition, cohort, phenotype, fixed and random effects, odds ratio, interval, donor support, multiplicity, stability and replication.
A donor or sample random intercept helps account for repeated cells from the same biological unit; it does not rescue a study with too few or perfectly confounded donors.
Keep the biological sample as the unit of replication. Many cells improve resolution, but they do not turn one donor, animal or culture into many independent experiments.
Section 27 of 36
27. Design an independent replication
Predefine the population and formula before analysing a new cohort. Matching tissue, timing and treatment context makes replication interpretable.
Fixed effects can adjust measured technical or biological covariates, but adjustment is credible only when the model is specified from subject knowledge rather than result hunting.
Use negative controls, label permutations or design-matched null analyses where valid. A trustworthy signal should exceed artefacts that the same workflow can manufacture.
Section 28 of 36
28. Separate association from clinical use
A reproducible disease association does not establish sensitivity, specificity, prospective prediction or treatment benefit. Those require dedicated validation designs.
The cellular subset must be defined before the association is interpreted. Cluster discovery and association testing on the same data can require extra validation against selection bias.
Report effect direction, uncertainty, sample support and sensitivity—not only a colourful embedding, selected feature or adjusted value.
Section 29 of 36
29. Connect MASC to school science
The method shows why repeated measurements from one source are related and why a fair comparison must represent several independent sources.
An odds ratio describes association with subset membership under the fitted model. It is not the same as absolute prevalence, risk, diagnostic accuracy or causal effect.
Create an evidence card naming the contrast, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 30 of 36
30. Try a classroom random-effect activity
Measure several leaves from each plant under two conditions, then compare a naive leaf-level conclusion with a plant-aware summary and discuss the difference.
The rheumatoid-arthritis application motivates MASC, but every new disease, tissue and assay needs its own sampling, quality control and replication evidence.
Write association, stratification, differential abundance or differential state when that is what was tested. Reserve cause, diagnosis and benefit for evidence that directly supports them.
Section 31 of 36
31. Connect to mathematics
Logistic regression, odds, random effects, confidence intervals and multiplicity turn a nested data structure into a disciplined claim.
MASC reverses the familiar viewpoint: the binary response is whether a cell belongs to a chosen subset, while the predictor of interest is a sample-level phenotype.
Archive sample manifests, preprocessing, feature roles, model formulae, random seeds, software versions and exclusions so another analyst can reconstruct the decision path.
Section 32 of 36
32. Connect to computing and AI literacy
A model can adjust declared variables, but it cannot discover every missing confounder or repair a study in which phenotype and batch are identical.
A donor or sample random intercept helps account for repeated cells from the same biological unit; it does not rescue a study with too few or perfectly confounded donors.
Keep the biological sample as the unit of replication. Many cells improve resolution, but they do not turn one donor, animal or culture into many independent experiments.
Section 33 of 36
33. Connect to school and career pathways
Students drawn to disease cohorts and quantitative evidence can explore medicine, immunology, epidemiology, statistics, public health or bioinformatics.
Fixed effects can adjust measured technical or biological covariates, but adjustment is credible only when the model is specified from subject knowledge rather than result hunting.
Use negative controls, label permutations or design-matched null analyses where valid. A trustworthy signal should exceed artefacts that the same workflow can manufacture.
Section 34 of 36
34. Create a family evidence habit
When a study names a disease-associated cell, ask how many people—not just cells—were studied and whether the population repeated in an independent cohort.
The cellular subset must be defined before the association is interpreted. Cluster discovery and association testing on the same data can require extra validation against selection bias.
Report effect direction, uncertainty, sample support and sensitivity—not only a colourful embedding, selected feature or adjusted value.
Section 35 of 36
35. Use precise MASC language
Say adjusted association with subset membership and donor-aware mixed-effects model. Avoid disease-causing cell, diagnostic marker or treatment target without direct evidence.
An odds ratio describes association with subset membership under the fitted model. It is not the same as absolute prevalence, risk, diagnostic accuracy or causal effect.
Create an evidence card naming the contrast, measured modality, statistical unit, model, result, validation status, unresolved alternatives and narrowest defensible claim.
Section 36 of 36
36. Finish with a cohort-aware bundle
Deliver participant manifests, subset definitions, formulae, convergence checks, corrected results, donor plots, leave-one-out analyses, independent replication and claim boundaries.
The rheumatoid-arthritis application motivates MASC, but every new disease, tissue and assay needs its own sampling, quality control and replication evidence.
Write association, stratification, differential abundance or differential state when that is what was tested. Reserve cause, diagnosis and benefit for evidence that directly supports them.
## Source date and scope note Primary papers 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