eduKateSG · Why Science?
Map local splice choices as a graph, then keep probability attached to every conclusion
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. MAJIQ detects and quantifies Local Splicing Variations (LSVs) from RNA-seq. It builds splice graphs from a genome annotation plus sample alignments, can include annotated and de novo junctions and retained introns, and represents uncertainty with probability distributions for PSI and changes in PSI. An LSV is local: its connected splice choices share a source or target exon, so it should not be mistaken automatically for a complete transcript. 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 MAJIQ project website; Current MAJIQ v3 documentation; MAJIQ primary study in eLife; MAJIQ v3 public-release announcement, 20 June 2025; 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 |
|---|---|---|---|
| Build | What splice connections exist locally? | GFF3 annotation plus SAM/BAM evidence | A sample-aware splice graph |
| Define | Which connections share a source or target? | Junctions and retained-intron connections | A local splicing variation |
| Quantify | How is probability distributed among choices? | Reads, priors and model settings | Posterior PSI distributions |
| Compare | How likely is a meaningful change? | Group models and delta-PSI threshold | Probability attached to a bounded local claim |
Inside this guide
1–12 · Foundations and design
- 1. Think in splice graphs
- 2. Define a local splicing variation
- 3. Keep local and full-length claims separate
- 4. Begin with consistent inputs
- 5. Build the splice graph
- 6. Discover de novo junctions
- 7. Represent retained introns
- 8. Identify the independent sample
- 9. Understand PSI inside an LSV
- 10. Understand delta-PSI
- 11. Keep the posterior distribution
- 12. Choose a meaningful-change threshold
13–24 · Models and estimation
- 13. Separate probability from certainty
- 14. Model groups appropriately
- 15. Use heterogeneity-aware questions
- 16. Interpret complex LSVs
- 17. Visualise with VOILA
- 18. Check read support
- 19. Check splice-site sequence
- 20. Check annotation sensitivity
- 21. Check build thresholds
- 22. Check sample influence
- 23. Check batch structure
- 24. Compare v3 and earlier formats carefully
25–36 · Diagnostics and interpretation
- 25. Use current version language precisely
- 26. Compare with LeafCutter fairly
- 27. Compare with SUPPA2 fairly
- 28. Connect to probability literacy
- 29. Connect to graph literacy
- 30. Connect to fair tests
- 31. Connect to education pathways
- 32. Use bounded language
- 33. Build a validation ladder
- 34. Audit one LSV end to end
- 35. Translate for a non-specialist
- 36. Did you know?
37–39 · Validation, learning and pathways
Section 1 of 39
1. Think in splice graphs
A splice graph represents exonic regions and the connections formed by splicing. Graph thinking is useful because one locus can offer branching routes rather than a single linear transcript.
Working checkpoint — splice graph: Draw nodes and edges for one simple exon-skipping example. 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 39
2. Define a local splicing variation
MAJIQ groups connections that share a common source or target exon into an LSV. The unit captures a local decision point, including binary and more complex alternatives.
Working checkpoint — local splicing variation: State whether the LSV is source- or target-defined. 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 39
3. Keep local and full-length claims separate
An LSV describes nearby competing connections. It does not necessarily reveal which distant splice choices occur together on the same RNA molecule.
Working checkpoint — junction connection: Use long-read or phased evidence for full-transcript 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 4 of 39
4. Begin with consistent inputs
Current MAJIQ documentation uses a GFF3 annotation with sample SAM or BAM alignments. Genome build, contig names and strand information must agree.
Working checkpoint — retained intron: Validate coordinate and chromosome-name consistency before building. 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 39
5. Build the splice graph
The build stage combines annotated structure with observed junction evidence across configured samples. The resulting graph becomes the reference for downstream quantification.
Working checkpoint — PSI posterior: Archive the configuration file, annotation and build logs. 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 39
6. Discover de novo junctions
MAJIQ can add junctions supported by alignments even when absent from the annotation. Discovery is valuable, but mapping artefacts and rare noise remain competing explanations.
Working checkpoint — delta-PSI probability: Require transparent read support and inspect decisive novel edges. 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 39
7. Represent retained introns
MAJIQ v3 documentation includes retained-intron connections in its splice-graph framework. Retention evidence behaves differently from a clean exon–exon junction and can be sensitive to pre-mRNA or coverage bias.
Working checkpoint — splice graph: Inspect intronic coverage and library preparation before claiming regulated retention. 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 39
8. Identify the independent sample
Biological replicates are independently sampled organisms, people, cultures or specimens. Read depth and technical lanes affect measurement, not the number of independent units.
Working checkpoint — local splicing variation: Make sample identity explicit in the configuration. 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 39
9. Understand PSI inside an LSV
Each connection receives a PSI value representing its share within that LSV. The PSI values across the competing connections sum to one under the model.
Working checkpoint — junction connection: Show all connection shares whenever discussing one edge. 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 39
10. Understand delta-PSI
Delta-PSI compares connection usage between conditions in a stated direction. Because LSV connection shares are constrained, their delta-PSI values sum to zero.
Working checkpoint — retained intron: Check the full vector before interpreting a single positive change. 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 39
11. Keep the posterior distribution
MAJIQ represents uncertainty probabilistically rather than reducing every result immediately to one point estimate. Broad and narrow posterior distributions communicate different levels of precision.
Working checkpoint — PSI posterior: Plot or summarise the posterior, not only its mean. 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 39
12. Choose a meaningful-change threshold
A probability such as Pr(|delta-PSI| greater than a threshold) depends on the threshold chosen. The threshold should reflect the scientific purpose, not be tuned after seeing results.
Working checkpoint — delta-PSI probability: Record the delta-PSI cutoff and justification before ranking LSVs. 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 39
13. Separate probability from certainty
A high posterior probability is conditional on data, model, annotation and settings. It is not an unconditional probability that a mechanism is true.
Working checkpoint — splice graph: Write the event, threshold and model beside the probability. 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 39
14. Model groups appropriately
Group comparisons should reflect independent replicates and the intended biological contrast. Heterogeneous groups can produce broad or multimodal patterns that a simple label obscures.
Working checkpoint — local splicing variation: Plot sample-level estimates before trusting a group summary. 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 39
15. Use heterogeneity-aware questions
MAJIQ tooling has evolved to address variation across samples and more than simple two-group summaries. The chosen analysis mode must match the question and current documentation.
Working checkpoint — junction connection: Name the exact command and inference mode rather than writing only “MAJIQ”. 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 39
16. Interpret complex LSVs
An LSV can contain more than two connections. Complex branching may better represent biology than forcing several disconnected binary events, but it also demands clearer visual explanation.
Working checkpoint — retained intron: Label every connection and its share in the figure. 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 39
17. Visualise with VOILA
VOILA is MAJIQ’s companion system for exploring LSVs, splice graphs and quantified changes. Interactive displays aid reasoning but do not replace exported, auditable evidence.
Working checkpoint — PSI posterior: Save static figures or reports for every headline LSV. 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 39
18. Check read support
Posterior estimates still depend on aligned reads. Low counts, mapping ambiguity or one influential sample can produce uncertain or misleading local patterns.
Working checkpoint — delta-PSI probability: Review junction counts and alignment quality under the graph. 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 39
19. Check splice-site sequence
Canonical motifs can add plausibility to de novo junctions, while non-canonical sites need particularly careful validation. Sequence context is supportive, not decisive.
Working checkpoint — splice graph: Inspect genome sequence and possible alignment alternatives. 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 39
20. Check annotation sensitivity
Changing GFF3 release can add exons, split structures or reclassify a connection from de novo to annotated. The graph and LSV set can change accordingly.
Working checkpoint — local splicing variation: Repeat key findings with one justified annotation sensitivity check. 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 39
21. Check build thresholds
Minimum read support and other build settings affect which edges enter the graph. A threshold can simplify noise while also excluding rare biology.
Working checkpoint — junction connection: Publish the settings and retained-edge 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 39
22. Check sample influence
Leave-one-replicate-out comparisons reveal LSVs whose posterior change depends on one specimen. This does not create more data; it exposes fragility.
Working checkpoint — retained intron: Report direction and probability stability across omissions. 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 39
23. Check batch structure
Library preparation, sequencing centre or processing date can align with the biological groups. No probabilistic model can identify an unobserved group effect separately from a perfectly confounded batch.
Working checkpoint — PSI posterior: Show the sample-by-batch table and mitigate at design time. 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 39
24. Compare v3 and earlier formats carefully
The official v3 announcement states that intermediate and final file formats are not compatible with v2. A reproducible migration should rebuild and document outputs instead of mixing versions.
Working checkpoint — delta-PSI probability: Keep version-specific environments and do not splice old binaries into new outputs. 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 39
25. Use current version language precisely
The current documentation checked for this guide displays a 3.0.11 development documentation version, while the public v3 release was announced on 20 June 2025.
Working checkpoint — splice graph: Record the actual installed release or commit, not the documentation label alone. 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 39
26. Compare with LeafCutter fairly
LeafCutter clusters observed introns; MAJIQ constructs LSVs in splice graphs and carries posterior PSI uncertainty. Overlapping loci do not imply identical hypotheses.
Working checkpoint — local splicing variation: Compare biological patterns after mapping each method’s unit. 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 39
27. Compare with SUPPA2 fairly
SUPPA2 derives event PSI from transcript abundance, whereas MAJIQ works from junction-oriented graph evidence. Disagreement can reveal annotation, quantification or unit-of-analysis differences.
Working checkpoint — junction connection: Align contrast and coordinates before calling one result a failure. 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 39
28. Connect to probability literacy
A probability distribution expresses a range of plausible values under assumptions. This offers a richer lesson than treating every scientific answer as one exact number.
Working checkpoint — retained intron: Ask learners how a wide interval should change confidence. 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 39
29. Connect to graph literacy
Nodes, edges, branches and local connectivity appear in transport maps, computer networks and splice graphs. The same representation can reveal alternate routes.
Working checkpoint — PSI posterior: Build a route-map analogy and then mark where it stops matching biology. 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 39
30. Connect to fair tests
Consistent inputs, independent repeats and a predeclared comparison remain the foundation beneath advanced Bayesian computation.
Working checkpoint — delta-PSI probability: Have students identify controlled variables before interpreting a posterior. 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 39
31. Connect to education pathways
The workflow connects molecular biology, statistics, graph algorithms and data visualisation. It can broaden awareness of STEM careers without guaranteeing a course place or job.
Working checkpoint — splice graph: Check official programme pages for current requirements. 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 39
32. Use bounded language
A MAJIQ result supports a local splicing difference with quantified posterior uncertainty. It does not alone prove a full isoform, protein consequence or causal phenotype.
Working checkpoint — local splicing variation: Keep “predicted local variation” separate from “validated mechanism”. 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 39
33. Build a validation ladder
Alignment inspection tests edges; targeted RNA assays test local structure; long reads test phasing; protein and phenotype assays test consequences.
Working checkpoint — junction connection: Choose the validation rung that directly challenges the headline. 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 39
34. Audit one LSV end to end
Trace one LSV from GFF3 and alignments through graph build, connection counts, posterior PSI, delta-PSI probability, VOILA view and annotation.
Working checkpoint — retained intron: An end-to-end audit reveals where each inference entered. 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 39
35. Translate for a non-specialist
A clear explanation names the local choices, estimated shift, probability threshold, independent sample count and remaining uncertainty.
Working checkpoint — PSI posterior: Avoid saying that probability proves the gene changed function. 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 39
36. Did you know?
Within an LSV, PSI values sum to one and delta-PSI values sum to zero. A gain on one connection must be balanced by losses elsewhere in that same local choice set.
Working checkpoint — delta-PSI probability: Use the constraint as a built-in consistency check. 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 39
37. Preserve computational provenance
MAJIQ v3 brought substantial changes to performance, storage and workflows. Reproducibility requires the executable version, config, reference files and exact commands.
Working checkpoint — splice graph: Archive machine-readable manifests with the result bundle. 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 39
38. Report uncertainty that resists a headline
A broad posterior or unstable graph is meaningful information about limited data. Do not force every locus into a binary changed/unchanged category.
Working checkpoint — local splicing variation: Keep unresolved LSVs visible for future measurement. 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 39
39. Finish with a reproducible bundle
Deliver GFF3, alignments or accessions, configuration, MAJIQ and VOILA versions, build logs, LSV outputs, posterior summaries, plots, sensitivity checks and bounded conclusions.
Working checkpoint — junction connection: A reproducible graph is as important as a persuasive picture. 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?
Contents · Previous section · Continue to the Science Learning Hub