eduKateSG · Why Science?
Let biological replicates shape the test—not merely decorate an average
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. rMATS is an event-based method for detecting differential alternative splicing from replicate RNA-seq. Its hierarchical modelling was designed to account for sampling uncertainty in individual samples and variability among biological replicates. The current rMATS turbo implementation accelerates the workflow and reports standard event classes using junction-only or junction-plus-exon evidence, but every result remains conditional on alignment, annotation, count thresholds and study design. 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; DRIMSeq and transcript usage; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official rMATS turbo repository and current v4.4.0 release information; rMATS primary study record in PubMed; Updated rMATS-turbo 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 |
|---|---|---|---|
| Count | Which reads support inclusion or skipping? | BAM alignments, GTF and read-length settings | Event-specific inclusion and skipping counts |
| Model | How variable is usage within each group? | Independent biological replicates | Replicate-aware inclusion estimates |
| Test | Does the stated contrast exceed the null difference? | Likelihood model and inclusion-difference threshold | Event-level p-value, FDR and effect |
| Validate | Is the event structurally and biologically supported? | Read views, sensitivity checks and orthogonal assays | A bounded, reproducible splice-event claim |
Inside this guide
1–12 · Foundations and design
- 1. Start with replicate-aware inference
- 2. Define the event catalogue
- 3. Recognise skipped exons
- 4. Recognise alternative splice sites
- 5. Recognise mutually exclusive exons
- 6. Recognise retained introns
- 7. Prepare compatible BAM files
- 8. Specify read length honestly
- 9. Keep group files exact
- 10. Identify independent replicates
- 11. Understand inclusion counts
- 12. Distinguish JC and JCEC outputs
13–24 · Models and estimation
- 13. Read inclusion level
- 14. Read inclusion-level difference
- 15. Account for sampling uncertainty
- 16. Account for between-replicate variability
- 17. Use the cstat threshold deliberately
- 18. Read p-values with FDR
- 19. Pair evidence with magnitude
- 20. Inspect effective-length assumptions
- 21. Inspect low-count events
- 22. Inspect one event in a browser
- 23. Check annotation sensitivity
- 24. Check aligner sensitivity
25–36 · Diagnostics and interpretation
- 25. Check sample influence
- 26. Check batch and pairing
- 27. Avoid double-counting related events
- 28. Compare JC and JCEC as a sensitivity
- 29. Use turbo without equating speed with truth
- 30. Compare with transcript-based PSI
- 31. Compare with graph-based methods
- 32. Connect to repeated measurements
- 33. Connect to ratios
- 34. Connect to computing
- 35. Connect to education pathways
- 36. Use bounded language
37–40 · Validation, learning and pathways
Section 1 of 40
1. Start with replicate-aware inference
The name rMATS expands to replicate Multivariate Analysis of Transcript Splicing. Its central lesson is that repeated biological samples reveal variability that a comparison of pooled means cannot.
Working checkpoint — replicate multivariate analysis: State the independent sample count before reporting event totals. 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 40
2. Define the event catalogue
rMATS uses an annotation to define testable alternative-splicing events. The catalogue depends on transcript structures, genome coordinates and the supplied GTF release.
Working checkpoint — inclusion level: Freeze the GTF and record how novel splice sites are handled. 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 40
3. Recognise skipped exons
A skipped-exon event contrasts reads supporting exon inclusion with reads supporting the junction that bypasses the exon. The event is local and does not identify a complete transcript.
Working checkpoint — junction count: Draw inclusion and skipping paths with coordinates. 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 40
4. Recognise alternative splice sites
Alternative 5-prime and 3-prime splice-site events compare junctions using different donor or acceptor positions. Coordinate convention and strand are essential to correct interpretation.
Working checkpoint — exon body count: Verify donor and acceptor labels on a known locus. 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 40
5. Recognise mutually exclusive exons
Mutually exclusive exon events represent alternative exon choices in an annotated structure. Short reads may not prove that the two exons never co-occur in every full-length molecule.
Working checkpoint — splicing event: Use full-length evidence for a strict exclusivity claim. 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 40
6. Recognise retained introns
Retained-intron events compare evidence for spliced and unspliced forms. Intronic signal can also reflect pre-mRNA, DNA contamination or uneven coverage.
Working checkpoint — sampling uncertainty: Inspect coverage and library protocol before naming regulated retention. 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 40
7. Prepare compatible BAM files
All samples should use the same genome build, aligner policy and coordinate conventions. Sorting, read pairing and strandedness settings must match the actual libraries.
Working checkpoint — replicate multivariate analysis: Run sample-level alignment and junction-quality checks first. 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 40
8. Specify read length honestly
rMATS uses read-length information in counting and effective-length calculations. Mixed or trimmed read lengths require settings consistent with the data and current documentation.
Working checkpoint — inclusion level: Record preprocessing and inspect length distributions. 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 40
9. Keep group files exact
The two group lists define the contrast and sample order. A swapped or duplicated BAM path creates a valid-looking analysis for the wrong experiment.
Working checkpoint — junction count: Resolve paths and compare them with the signed sample sheet. 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 40
10. Identify independent replicates
The biological unit is the person, organism, culture or specimen independently sampled or assigned. Technical lanes should be combined or modelled appropriately, not counted as new organisms.
Working checkpoint — exon body count: Make pairing and repeated measures explicit. 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 40
11. Understand inclusion counts
Inclusion-supporting evidence can include junctions linking the alternative exon to its neighbours. Skipping evidence comes from the bypass junction for a skipped-exon event.
Working checkpoint — splicing event: Show the actual count columns for representative events. 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 40
12. Distinguish JC and JCEC outputs
rMATS reports junction-count-only results and results combining junction counts with exon-body counts. The evidence sources and possible biases differ.
Working checkpoint — sampling uncertainty: Name the chosen output type in every result table. 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 40
13. Read inclusion level
The inclusion level is an estimated proportion derived from event-specific counts and effective lengths. It is not a direct census of molecules or cells.
Working checkpoint — replicate multivariate analysis: Report sample-level inclusion values with the formula context. 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 40
14. Read inclusion-level difference
The reported difference follows the order of group one minus group two. Its sign changes if group labels are reversed.
Working checkpoint — inclusion level: Write the group order in the heading and figure legend. 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 40
15. Account for sampling uncertainty
Limited reads create uncertainty within each replicate. rMATS modelling was designed to account for that sampling uncertainty rather than treat an estimated proportion as exact.
Working checkpoint — junction count: Inspect events with sparse counts even when their estimated difference is large. 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 40
16. Account for between-replicate variability
Biological samples rarely have identical inclusion levels. A credible group claim must consider how much samples vary within each condition.
Working checkpoint — exon body count: Plot all replicates and do not hide dispersion behind bars. 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 40
17. Use the cstat threshold deliberately
rMATS can test against a user-specified cutoff for the inclusion-level difference rather than only a zero difference. This changes the null hypothesis and interpretation.
Working checkpoint — splicing event: Record cstat and justify it before reviewing results. 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 40
18. Read p-values with FDR
Genome-wide event testing creates many comparisons. False-discovery-rate adjustment applies to the tested family in each result table under its filters.
Working checkpoint — sampling uncertainty: Keep raw, adjusted and complete result sets together. 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 40
19. Pair evidence with magnitude
A low FDR does not tell readers whether the inclusion shift is large enough to matter. A large shift with sparse reads may remain unreliable.
Working checkpoint — replicate multivariate analysis: Report inclusion difference, counts, sample pattern and FDR together. 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 40
20. Inspect effective-length assumptions
Read length and event geometry affect how many positions could generate inclusion or skipping evidence. Effective-length correction is part of the estimate, not an optional decoration.
Working checkpoint — inclusion level: Audit one event’s lengths and count normalisation. 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 40
21. Inspect low-count events
Zeros and tiny counts can produce extreme proportions. Filtering protects interpretation but also defines which events enter the testing family.
Working checkpoint — junction count: Publish filtering thresholds and before–after counts. 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 40
22. Inspect one event in a browser
A genome browser can reveal soft clipping, repetitive mapping, neighbouring events or coverage gaps that summary tables conceal.
Working checkpoint — exon body count: Review every event used for a strong biological headline. 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 40
23. Check annotation sensitivity
A newer or alternative GTF can change exon boundaries and event membership. This may alter counts and the multiplicity family.
Working checkpoint — splicing event: Run one planned annotation sensitivity for key claims. 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 40
24. Check aligner sensitivity
Splice-aware aligners and parameters differ in novel-junction handling and multimapping. A favourite result that vanishes under a justified alternative deserves caution.
Working checkpoint — sampling uncertainty: Keep both workflows comparable and report disagreements. 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 40
25. Check sample influence
Leave-one-replicate-out runs reveal events driven by a single sample. Repeated refitting is a diagnostic, not a source of extra independent evidence.
Working checkpoint — replicate multivariate analysis: Flag sign changes and unstable FDR status. 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 40
26. Check batch and pairing
Batch effects and paired designs require appropriate handling in study design and supported model settings. Complete confounding cannot be repaired by adding a label.
Working checkpoint — inclusion level: Show the sample structure before interpreting contrasts. 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 40
27. Avoid double-counting related events
One structural change can generate several event rows in the same gene. Counting each row as an independent biological mechanism exaggerates discovery.
Working checkpoint — junction count: Group overlapping events and inspect their shared junctions. 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 40
28. Compare JC and JCEC as a sensitivity
Agreement can strengthen confidence; disagreement may arise from exon-body coverage, overlapping expression or low junction support. Neither output wins automatically.
Working checkpoint — exon body count: Explain the evidence difference for headline events. 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 40
29. Use turbo without equating speed with truth
rMATS turbo is a faster C/Cython reimplementation and the official repository checked here identifies v4.4.0. Computational efficiency enables scale but does not fix confounding or poor annotation.
Working checkpoint — splicing event: Report performance and inferential validity separately. 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 40
30. Compare with transcript-based PSI
SUPPA2 estimates event PSI from transcript abundance, while rMATS counts event-supporting reads directly. The methods can disagree because their evidence and models differ.
Working checkpoint — sampling uncertainty: Map coordinates and event definitions before comparison. 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 40
31. Compare with graph-based methods
MAJIQ LSVs and LeafCutter clusters can represent complex local choices that do not correspond one-to-one with standard rMATS events.
Working checkpoint — replicate multivariate analysis: Compare loci and splice paths, not row counts alone. 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 40
32. Connect to repeated measurements
Replicates show how much natural variation occurs when a measurement is repeated on independent units. This is the same reason school experiments need more than one trial.
Working checkpoint — inclusion level: Ask which repeats are biological and which are merely technical. 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 40
33. Connect to ratios
Inclusion level is a proportion built from competing evidence and effective lengths. Fractions remain interpretable only when the denominator is visible.
Working checkpoint — junction count: Reconstruct a toy inclusion estimate with counters. 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 40
34. Connect to computing
File lists, genomic coordinates, parallel jobs and structured output tables are part of the scientific argument. Automation should make identity checks stricter, not looser.
Working checkpoint — exon body count: Stop the pipeline on duplicated samples or mismatched contigs. 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 40
35. Connect to education pathways
Differential-splicing analysis blends biology, statistics and programming and can illustrate work in genomics or bioinformatics.
Working checkpoint — splicing event: Use official current sources for admissions and careers; a tool is not a credential. 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 40
36. Use bounded language
An rMATS event supports differential inclusion under a specified model and input set. It does not alone prove a full isoform, protein change, diagnosis or causal mechanism.
Working checkpoint — sampling uncertainty: Use the narrowest verb supported by the evidence. 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 37 of 40
37. Did you know?
Two groups can have the same mean inclusion level yet very different replicate-to-replicate spread. A replicate-aware method keeps that spread in the scientific question.
Working checkpoint — replicate multivariate analysis: Plot points before trusting an average. 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 38 of 40
38. Build a validation ladder
Browser review checks mapping; targeted RNA assays check the local event; long reads test full structures; protein and phenotype assays test consequences.
Working checkpoint — inclusion level: Choose validation according to the claim being advanced. 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 39 of 40
39. Audit one event end to end
Trace one event from GTF through inclusion and skipping reads, effective lengths, sample estimates, group model, FDR, sensitivity checks and final wording.
Working checkpoint — junction count: An audit should reproduce both the number and its meaning. 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 40 of 40
40. Finish with a reproducible bundle
Deliver GTF, BAM accessions, sample lists, command, version, read settings, JC and JCEC tables, filters, plots, diagnostics, sensitivity results and bounded conclusion.
Working checkpoint — exon body count: A complete archive lets another reader challenge the strongest claim fairly. 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.
Contents · Previous section · Continue to the Science Learning Hub