eduKateSG · Why Science?
Model whether a transcript is detected and how strongly it is expressed when present—two questions that reveal different parts of sparse single-cell evidence
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 Muscat Multi Sample Multi Group Differential State Analysis; Why Science Scdd Distributional Changes Single Cell Expression; Why Science Diffcyt High Resolution Clustering Differential Discovery; 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: MAST primary study; MAST PubMed record; MAST open full text; Official Bioconductor MAST 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.
MAST—Model-based Analysis of Single-cell Transcriptomics—was introduced by Finak and colleagues in Genome Biology in 2015. Its two-part hurdle model represents both whether expression is detected in a cell and the continuous expression level when detected, while permitting covariate adjustment such as cellular detection rate. The framework is useful for differential expression and gene-set questions in sparse single-cell data, but reliable biological claims still depend on experimental units, specimen-aware validation, model diagnostics and a clear boundary between cellular association and organism-level inference.
Inside this guide
1–12 · Foundations and models
- 1. Begin with many zeros
- 2. Ask two questions at once
- 3. Understand the hurdle
- 4. Name the scientific unit
- 5. Define the contrast
- 6. Keep the scale visible
- 7. Read the 2015 evidence in scope
- 8. Build a cell-and-sample manifest
- 9. Filter genes without outcome chasing
- 10. Inspect cellular detection rate
- 11. Choose covariates from the design
- 12. Check cell-population identity
13–24 · Evidence, testing and applications
- 13. Fit the discrete component
- 14. Fit the continuous component
- 15. Practise with a fictional MAST table
- 16. Combine evidence carefully
- 17. Read coefficients on their scale
- 18. Adjust for many genes
- 19. Plot both components
- 20. Check residuals and convergence
- 21. Resample by specimen
- 22. Compare cell-level and pseudobulk results
- 23. Challenge pseudo-replication
- 24. Challenge detection as biology
25–36 · Learning, decisions and pathways
- 25. Compare with muscat
- 26. Compare with scDD
- 27. Build a hurdle-model evidence card
- 28. Design orthogonal confirmation
- 29. Connect MAST to school science
- 30. Try a zero-and-positive 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 MAST language
- 36. Finish with a two-part evidence bundle
Section 1 of 36
1. Begin with many zeros
Single-cell expression matrices contain numerous zero or undetected values. Some genes are inactive in a cell; other transcripts are present but missed by limited sampling and assay efficiency.
A hurdle model has a discrete part for whether expression is detected and a continuous part for the positive measurements, allowing two biologically different patterns to be separated.
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. Ask two questions at once
For each gene, ask whether detection probability differs between groups and whether the positive expression level differs among cells where the transcript was detected.
The cellular detection rate—the fraction of genes detected in a cell—can capture technical and biological variation and is often included as a covariate in MAST analyses.
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. Understand the hurdle
The model treats zero versus non-zero expression as one process and positive expression as another. The two components meet at a hurdle rather than being forced into one average.
A gene can differ because more cells cross the detection threshold, because positive cells express more RNA, or because both components change.
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 scientific unit
Cells are observations, yet donors or biological samples define replication for population claims. Record both levels and choose inference that respects their relationship.
Cell-level precision does not remove donor dependence. Multi-sample studies should preserve specimen structure through aggregation, random effects or specimen-aware resampling suited to the question.
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. Define the contrast
State the cell population, condition, time point and coefficient of interest before fitting. A long ranked list is useful only when its comparison is unambiguous.
MAST can combine component evidence and support gene-set testing, but pathway interpretation should remain conditional on the tested gene universe and correlation structure.
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. Keep the scale visible
Document whether values are counts, normalised counts or log-transformed expression and how positive values enter the continuous component.
Zeros in single-cell counts can reflect biology, sampling and measurement. The model describes their statistical pattern; it does not label every zero as one mechanism.
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 2015 evidence in scope
Finak and colleagues introduced MAST as a two-part regression framework for single-cell transcriptomic data, using detection and positive-expression components with covariate adjustment and gene-set extensions.
A hurdle model has a discrete part for whether expression is detected and a continuous part for the positive measurements, allowing two biologically different patterns to be separated.
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 a cell-and-sample manifest
Map every barcode to donor, condition, batch, tissue, cell annotation and quality metrics. Check that group and batch are not perfectly aligned.
The cellular detection rate—the fraction of genes detected in a cell—can capture technical and biological variation and is often included as a covariate in MAST analyses.
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. Filter genes without outcome chasing
Use a declared prevalence or count threshold so genes with almost no information do not consume the testing budget or produce unstable coefficients.
A gene can differ because more cells cross the detection threshold, because positive cells express more RNA, or because both components change.
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. Inspect cellular detection rate
Plot the fraction of genes detected per cell across samples, batches and populations. Decide whether it is a nuisance proxy, biological feature or both in this experiment.
Cell-level precision does not remove donor dependence. Multi-sample studies should preserve specimen structure through aggregation, random effects or specimen-aware resampling suited to the question.
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. Choose covariates from the design
Include variables justified before results are viewed, such as batch or stimulation. Too many terms can overfit a small number of biological samples.
MAST can combine component evidence and support gene-set testing, but pathway interpretation should remain conditional on the tested gene universe and correlation structure.
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. Check cell-population identity
Differential expression inside a cluster is interpretable only if the compared cells represent a coherent and comparable population across samples.
Zeros in single-cell counts can reflect biology, sampling and measurement. The model describes their statistical pattern; it does not label every zero as one mechanism.
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. Fit the discrete component
Model whether the gene is detected in each cell. A positive group coefficient suggests a higher probability of crossing the observed detection threshold under the fitted design.
A hurdle model has a discrete part for whether expression is detected and a continuous part for the positive measurements, allowing two biologically different patterns to be separated.
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. Fit the continuous component
Among detected cells, model expression magnitude. A coefficient here describes positive measurements, not the entire cell population by itself.
The cellular detection rate—the fraction of genes detected in a cell—can capture technical and biological variation and is often included as a covariate in MAST analyses.
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 MAST table
This teaching table is invented and is not a result from the MAST paper.
| Fictional gene | Detection change | Positive-expression change | Combined reading | Donors |
|---|---|---|---|---|
| GENE-X | Higher | Similar | More cells detected | 10 |
| GENE-Y | Similar | Higher | Stronger when present | 10 |
| GENE-Z | Higher | Lower | Mixed pattern | 4 |
A gene can differ because more cells cross the detection threshold, because positive cells express more RNA, or because both components change.
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. Combine evidence carefully
A joint test can ask whether either component differs, but the follow-up explanation should still show which component carries the signal and in what direction.
Cell-level precision does not remove donor dependence. Multi-sample studies should preserve specimen structure through aggregation, random effects or specimen-aware resampling suited to the question.
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. Read coefficients on their scale
Log-odds from the discrete model and continuous-expression coefficients answer different questions. Translate them separately rather than adding them into an intuitive but incorrect percentage.
MAST can combine component evidence and support gene-set testing, but pathway interpretation should remain conditional on the tested gene universe and correlation structure.
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 for many genes
Declare the tested gene universe and false-discovery procedure. Filtering, gene-set selection and repeated subgroup analyses all shape the multiplicity family.
Zeros in single-cell counts can reflect biology, sampling and measurement. The model describes their statistical pattern; it does not label every zero as one mechanism.
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. Plot both components
Show detection proportions and positive-expression distributions per biological sample. A combined test statistic alone can hide opposing components or one influential specimen.
A hurdle model has a discrete part for whether expression is detected and a continuous part for the positive measurements, allowing two biologically different patterns to be separated.
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. Check residuals and convergence
Inspect model warnings, fitted probabilities, residual patterns and extreme coefficients. Separation or sparse positive values can make apparently decisive estimates unstable.
The cellular detection rate—the fraction of genes detected in a cell—can capture technical and biological variation and is often included as a covariate in MAST analyses.
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. Resample by specimen
When several biological samples are present, bootstrap or cross-validate at the donor level rather than scattering cells from one donor across training and test folds.
A gene can differ because more cells cross the detection threshold, because positive cells express more RNA, or because both components change.
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. Compare cell-level and pseudobulk results
Disagreement may reveal sensitivity to zero modelling, within-sample weighting or low donor count. Treat the comparison as a diagnostic, not a contest with one automatic winner.
Cell-level precision does not remove donor dependence. Multi-sample studies should preserve specimen structure through aggregation, random effects or specimen-aware resampling suited to the question.
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. Challenge pseudo-replication
A million cells from two donors cannot support the same population claim as balanced evidence from many donors. Report donor counts prominently beside cell counts.
MAST can combine component evidence and support gene-set testing, but pathway interpretation should remain conditional on the tested gene universe and correlation structure.
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. Challenge detection as biology
Higher detection may reflect RNA content, sequencing depth, cell size or state. Covariate adjustment helps only for measured, correctly specified factors.
Zeros in single-cell counts can reflect biology, sampling and measurement. The model describes their statistical pattern; it does not label every zero as one mechanism.
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 muscat
muscat organises multi-sample, multi-group differential-state analysis and pseudobulk comparisons. MAST supplies a cell-level hurdle model focused on detection and positive expression.
A hurdle model has a discrete part for whether expression is detected and a continuous part for the positive measurements, allowing two biologically different patterns to be separated.
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 scDD
scDD models full distributional patterns and mixture changes across two conditions. MAST decomposes expression into detection and positive-value regression components.
The cellular detection rate—the fraction of genes detected in a cell—can capture technical and biological variation and is often included as a covariate in MAST analyses.
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 hurdle-model evidence card
Record gene, cell population, discrete and continuous estimates, joint evidence, covariates, donor support, diagnostics, multiplicity and validation status.
A gene can differ because more cells cross the detection threshold, because positive cells express more RNA, or because both components change.
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 orthogonal confirmation
Validate detection frequency and positive magnitude with targeted sequencing, RNA imaging, flow-based protein measurement or another assay appropriate to the mechanism.
Cell-level precision does not remove donor dependence. Multi-sample studies should preserve specimen structure through aggregation, random effects or specimen-aware resampling suited to the question.
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 MAST to school science
A familiar class result can have two parts: how many students attempted a question and how well the attempts scored. Averaging alone can blur the difference.
MAST can combine component evidence and support gene-set testing, but pathway interpretation should remain conditional on the tested gene universe and correlation structure.
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 zero-and-positive activity
Use bags containing blank and numbered cards. Compare groups by the chance of drawing a number and by the size of numbers drawn, then combine the interpretations.
Zeros in single-cell counts can reflect biology, sampling and measurement. The model describes their statistical pattern; it does not label every zero as one mechanism.
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
Binary outcomes, conditional means, regression coefficients, covariates and multiple testing demonstrate how one dataset can contain several linked questions.
A hurdle model has a discrete part for whether expression is detected and a continuous part for the positive measurements, allowing two biologically different patterns to be separated.
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
Software can fit both components rapidly, but it cannot decide whether an undetected transcript is biological silence or measurement failure for an individual cell.
The cellular detection rate—the fraction of genes detected in a cell—can capture technical and biological variation and is often included as a covariate in MAST analyses.
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
Curiosity about sparse data can open routes into biology, mathematics, statistics, data science, laboratory medicine or computational genomics.
A gene can differ because more cells cross the detection threshold, because positive cells express more RNA, or because both components change.
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 study says a gene increased, ask whether more cells expressed it, positive cells expressed more, or both—and how many independent samples agreed.
Cell-level precision does not remove donor dependence. Multi-sample studies should preserve specimen structure through aggregation, random effects or specimen-aware resampling suited to the question.
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 MAST language
Say higher detection probability, altered positive-expression level or joint differential expression under the fitted model. Avoid translating a cell-level association directly into patient benefit.
MAST can combine component evidence and support gene-set testing, but pathway interpretation should remain conditional on the tested gene universe and correlation structure.
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 two-part evidence bundle
Deliver manifests, transformation rules, filters, covariates, component estimates, diagnostics, donor-aware plots, multiplicity control, sensitivity checks and orthogonal validation.
Zeros in single-cell counts can reflect biology, sampling and measurement. The model describes their statistical pattern; it does not label every zero as one mechanism.
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