eduKateSG · Why Science?
Screen broadly, confirm precisely: why the order of statistical questions matters
Choose the closest route, then return to the full index whenever you need the wider evidence chain.
Science learning becomes powerful when students can tell a measured pattern from a model, a statistical decision from a mechanism, and a useful lead from a finished conclusion. stageR is a post-processing framework for stage-wise analysis. It first screens an aggregate unit such as a gene, then confirms which lower-level hypotheses—such as transcripts—carry evidence inside selected genes. The scientific value is not a shortcut to significance: it is an explicit hierarchy that matches a broad biological question to a narrower follow-up question. This guide supports Primary Science, PSLE Science, Secondary Science, O-Level Science and STEM education as a longform evidence-reading route; it is not medical advice, a laboratory protocol, an admission promise or a career guarantee.
Related eduKate reading: DEXSeq and differential exon usage; EventPointer and interpretable splicing events; Swish and inferential replicates; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official Bioconductor stageR package, release 3.23; stageR primary study in Genome Biology; stageR primary study record in PubMed; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science teaching and learning syllabus; 2026 MOE G2 Computing syllabus. Package interfaces and recommendations can change, so reproduce an analysis only with the documentation for the installed release.
| Stage | Question | Evidence | Responsible output |
|---|---|---|---|
| Screen | Is this gene worth opening? | One gene-level p-value | A selected gene, subject to the chosen screening rule |
| Confirm | Which transcripts support the signal? | Transcript-level p-values mapped to genes | Adjusted confirmation evidence within selected genes |
| Report | What family of errors is controlled? | Method, alpha and mapping record | A bounded gene-level false-discovery statement |
| Validate | Does the pattern survive scrutiny? | Plots, sensitivity checks and independent evidence | A candidate for follow-up, not a proved mechanism |
Inside this guide
1–12 · Foundations and design
- 1. Start with the hierarchy
- 2. Separate screening from confirmation
- 3. Name the error rate exactly
- 4. Use one screening value per gene
- 5. Map confirmation rows deliberately
- 6. Keep biological replicates independent
- 7. Choose compatible upstream tests
- 8. Inspect null calibration upstream
- 9. Declare the tested family
- 10. Practise with a card-and-folder model
- 11. Build the pScreen object carefully
- 12. Build the pConfirmation matrix
13–24 · Models and estimation
- 13. Set the stage-wise method explicitly
- 14. Run screening before interpretation
- 15. Read adjusted confirmation values
- 16. Distinguish gene selection from transcript evidence
- 17. Plot the underlying measurements
- 18. Test the annotation map
- 19. Test filtering sensitivity
- 20. Test sample influence
- 21. Compare a flat analysis honestly
- 22. Connect to fair testing
- 23. Connect to mathematics
- 24. Connect to computing
25–36 · Diagnostics and interpretation
- 25. Connect to education pathways
- 26. Ask the parent question
- 27. Use precise language
- 28. Archive the decision path
- 29. Audit the selected-gene count
- 30. Check one complete gene by hand
- 31. Report genes with no confirmations
- 32. Separate discovery from prioritisation
- 33. Plan an independent replication
- 34. Teach conditional decisions
- 35. Did you know?
- 36. Finish with a reproducible bundle
Section 1 of 36
1. Start with the hierarchy
A transcript belongs to a gene, so the hypotheses are naturally nested. stageR uses that structure instead of treating every transcript result as an unrelated line in one enormous table.
Working checkpoint — screening hypothesis: Write the gene-level and transcript-level scientific questions in plain language before importing any p-values. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 2 of 36
2. Separate screening from confirmation
The screening stage asks whether an aggregate unit shows enough evidence to justify a closer look. The confirmation stage asks which component hypotheses explain that selection.
Working checkpoint — confirmation hypothesis: Keep the two input matrices and their identifiers separate so a later audit can reconstruct each stage. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 3 of 36
3. Name the error rate exactly
The primary paper frames control at the gene level for a stage-wise procedure. That statement is different from claiming a conventional transcript-level false discovery rate across every transcript.
Working checkpoint — overall false discovery rate: Copy the exact target error rate and alpha into the methods record rather than writing only “FDR controlled”. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 4 of 36
4. Use one screening value per gene
A valid screening table has one value for each gene or other aggregate unit. Duplicate gene rows, mixed identifier versions or silent many-to-many joins can corrupt the hierarchy before testing begins.
Working checkpoint — adjusted confirmation p-value: Count rows before and after every join, and preserve an unmapped-identifier report. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 5 of 36
5. Map confirmation rows deliberately
Transcript confirmation values must map to the correct parent gene. Version suffixes, retired annotations and duplicated transcript identifiers are not cosmetic details.
Working checkpoint — gene–transcript map: Freeze the annotation release and save the exact transcript-to-gene table used by the analysis. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 6 of 36
6. Keep biological replicates independent
stageR reorganises statistical decisions; it cannot repair pseudoreplication. Repeated reads, technical lanes or repeated software runs do not become new biological samples.
Working checkpoint — independent biological replicate: Record the organism, person, culture or specimen that was independently assigned or observed. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 7 of 36
7. Choose compatible upstream tests
The screening and confirmation p-values normally come from upstream differential-expression or differential-usage methods. Their designs, contrasts and filtering rules must describe the same comparison.
Working checkpoint — screening hypothesis: Reject a workflow where the gene screen and transcript confirmation use incompatible samples or factor codings. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 8 of 36
8. Inspect null calibration upstream
A stage-wise wrapper assumes the supplied p-values carry meaningful calibration. Confounding, model misspecification or a poor count model can pass through a beautifully organised hierarchy.
Working checkpoint — confirmation hypothesis: Use diagnostics from the upstream method and inspect p-value behaviour before stageR. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 9 of 36
9. Declare the tested family
Filtering changes the family of hypotheses. A transparent analysis states which genes and transcripts entered screening, which were excluded and why.
Working checkpoint — overall false discovery rate: Save counts of features at import, after filtering, after mapping and after testing. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 10 of 36
10. Practise with a card-and-folder model
Imagine each gene as a folder and each transcript as a card inside it. Opening every folder invites many accidental highlights; screening chooses folders first, then confirmation examines cards inside them.
Working checkpoint — adjusted confirmation p-value: Use coloured classroom cards to model the logic without pretending that the activity reproduces RNA-seq statistics. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 11 of 36
11. Build the pScreen object carefully
The screening vector should have names that identify aggregate hypotheses. Missing names or recycled row order can create a valid-looking object tied to the wrong biology.
Working checkpoint — gene–transcript map: Check identity, not only dimension: compare sorted names and test a few known mappings by hand. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 12 of 36
12. Build the pConfirmation matrix
Confirmation evidence is commonly arranged with component hypotheses associated with each gene. Padding, missing values and unequal transcript counts require deliberate handling.
Working checkpoint — independent biological replicate: Document how genes with one transcript, missing p-values or filtered transcripts are represented. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 13 of 36
13. Set the stage-wise method explicitly
Software defaults are convenient but not a methods section. Record the package version, constructor arguments, selected method and target alpha.
Working checkpoint — screening hypothesis: Use the current Bioconductor vignette for the installed release, not an undated code fragment copied from elsewhere. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 14 of 36
14. Run screening before interpretation
The first-stage decision limits where confirmation claims may be made. Do not browse transcript rankings first and then retrofit a screen around favourite examples.
Working checkpoint — confirmation hypothesis: Lock the analysis plan and contrast before viewing biological labels attached to the smallest p-values. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 15 of 36
15. Read adjusted confirmation values
A small raw transcript p-value is not automatically a confirmed discovery. The stage-wise adjustment reflects the hierarchy and the number of selected families.
Working checkpoint — overall false discovery rate: Report the adjusted confirmation value used for the decision, not a convenient raw value from another table. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 16 of 36
16. Distinguish gene selection from transcript evidence
A selected gene says the screen passed. A confirmed transcript says a specified lower-level hypothesis passed its confirmation rule. Neither statement alone identifies a molecular mechanism.
Working checkpoint — adjusted confirmation p-value: Use verbs such as “selected” and “confirmed under the stage-wise procedure” rather than “proved”. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 17 of 36
17. Plot the underlying measurements
Hierarchical adjusted values do not replace sample-level plots. A signal driven by one sample, near-zero abundance or unstable quantification still needs scrutiny.
Working checkpoint — gene–transcript map: Plot every biological replicate and show denominators or gene totals when the question concerns usage. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 18 of 36
18. Test the annotation map
Transcript catalogues change. Repeating the map with a justified alternative release can reveal whether discoveries depend on annotation boundaries or identifier retirement.
Working checkpoint — independent biological replicate: Treat annotation sensitivity as evidence about robustness, not as a way to shop for more discoveries. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 19 of 36
19. Test filtering sensitivity
Reasonable abundance filters can change both upstream p-values and the multiplicity structure. Plan a small set of defensible filters and compare conclusions.
Working checkpoint — screening hypothesis: Record all tested filters, including ones that weaken the preferred result. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 20 of 36
20. Test sample influence
Leave-one-sample-out checks can reveal whether selection depends on a single influential observation. They do not replace replication, but they make fragility visible.
Working checkpoint — confirmation hypothesis: Flag genes whose screen or confirmation status changes repeatedly when one independent sample is removed. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 21 of 36
21. Compare a flat analysis honestly
A flat transcript-level adjustment answers a different multiplicity question. Compare it only after aligning inputs, contrasts, filters and effect scales.
Working checkpoint — overall false discovery rate: Explain why results differ instead of calling one list correct by default. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 22 of 36
22. Connect to fair testing
The same habits appear in school science: define the broad question, decide the focused follow-up, control comparisons and keep observations separate from explanations.
Working checkpoint — adjusted confirmation p-value: For Primary Science, PSLE Science and Secondary Science, practise naming the question before reading the answer choices. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 23 of 36
23. Connect to mathematics
Sets, conditional decisions, probabilities and error rates make the hierarchy inspectable. The lesson is that a percentage is meaningful only when its denominator and decision family are known.
Working checkpoint — gene–transcript map: Ask students to identify the denominator behind every reported rate. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 24 of 36
24. Connect to computing
Identifiers, joins, matrices and reproducible scripts are part of the scientific argument. A correct formula applied to misaligned rows produces confidently wrong output.
Working checkpoint — independent biological replicate: Use assertions that stop the script when gene and transcript names fail to match. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 25 of 36
25. Connect to education pathways
This workflow links biology, statistics and computing. It can illuminate routes into life science, biotechnology, bioinformatics and data science without promising admission or employment.
Working checkpoint — screening hypothesis: Use official course and institution pages when making any later pathway decision. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 26 of 36
26. Ask the parent question
A useful family question is: what was screened first, what was confirmed second, and which error rate does the headline actually describe?
Working checkpoint — confirmation hypothesis: That single question often reveals whether a complicated result has been communicated responsibly. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 27 of 36
27. Use precise language
Say that stageR supports stage-wise screening and confirmation under the stated inputs and assumptions. Avoid turning adjusted evidence into causal or clinical certainty.
Working checkpoint — overall false discovery rate: Reserve mechanism, diagnosis and benefit for experiments and validation capable of supporting those claims. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 28 of 36
28. Archive the decision path
Keep upstream result objects, mappings, filters, code, package versions, alpha, complete outputs and plots. A shortlist without its decision path cannot be audited.
Working checkpoint — adjusted confirmation p-value: Use stable accessions or data dictionaries where raw data cannot be redistributed. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 29 of 36
29. Audit the selected-gene count
The number of genes entering confirmation depends on the screening threshold and upstream calibration. An unexpectedly large or tiny selected set is a diagnostic clue, not merely a publishing inconvenience.
Working checkpoint — gene–transcript map: Compare the selected proportion with p-value histograms, effect distributions and known design power before interpretation. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 30 of 36
30. Check one complete gene by hand
Choose a gene with several transcripts and trace every value from the upstream files through screening, mapping, confirmation adjustment and the final table.
Working checkpoint — independent biological replicate: A hand audit catches row-order errors that summary plots can miss. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 31 of 36
31. Report genes with no confirmations
A gene may pass the screen while no transcript passes confirmation. That result is informative about diffuse evidence, limited power or incompatibility between aggregate and component tests.
Working checkpoint — screening hypothesis: Do not erase these genes from the narrative just because they lack a neat component label. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 32 of 36
32. Separate discovery from prioritisation
The stage-wise decision controls a defined error rate; later ranking by effect size, annotation or feasibility is a different prioritisation step.
Working checkpoint — confirmation hypothesis: Preserve the statistical status while adding transparent biological priorities in separate columns. Read the numerical result together with sample-level plots, effect size and a sensitivity check capable of weakening the headline.
For future study, connect the idea to biology, mathematics and computing while checking current official pathway information separately.
Section 33 of 36
33. Plan an independent replication
A new cohort, perturbation or targeted assay can test whether a selected gene and confirmed transcript pattern generalises beyond the discovery data.
Working checkpoint — overall false discovery rate: Replication should repeat the scientific contrast, not merely reuse the same counts with another package. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 34 of 36
34. Teach conditional decisions
The confirmation question is asked only after the parent gene passes screening. This conditional sequence is a practical way to show why decision trees need explicit entry rules.
Working checkpoint — adjusted confirmation p-value: Have students write the “if” statement before calculating any follow-up comparison. Separate what the model estimates from the biological story proposed afterward, then name the extra evidence that story would require.
For students, the transferable habit is to ask: What was measured? Compared with what? What alternative explanation remains?
Section 35 of 36
35. Did you know?
A two-stage design can be more scientifically readable because it mirrors the questions people naturally ask: first “is this gene involved?”, then “which transcript carries the evidence?”
Working checkpoint — gene–transcript map: The gain comes from matching the hierarchy—not from running the same test twice. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 36 of 36
36. Finish with a reproducible bundle
Deliver the screening vector, confirmation matrix, gene–transcript map, analysis script, session information, complete adjusted results, diagnostics and the narrowest defensible interpretation.
Working checkpoint — independent biological replicate: A complete negative result is more useful than a selective success story with missing rows. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Contents · Previous section · Continue to the Science Learning Hub