eduKateSG · Why Science?
Predict local splice alternatives from aligned reads, then quantify the paths that actually cross their boundaries
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. SGSeq is an R/Bioconductor package for predicting and quantifying splice events from genome-aligned RNA-seq data. It derives exon and splice-junction features, represents them as splice graphs, identifies local alternative paths between shared start and end boundaries and estimates relative variant usage from compatible reads. The event-centric view can represent unannotated and complex alternatives without claiming that short reads reveal every 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: SplAdder and splicing-graph event discovery; MAJIQ and local splicing variations; JunctionSeq and exon–junction usage; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official Bioconductor SGSeq package, release 3.23 and version 1.46.0; Official SGSeq release vignette; SGSeq primary study full text; SGSeq 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 |
|---|---|---|---|
| Predict | Which exons and junctions are supported? | Genome-aligned reads plus optional annotation | A transcript-feature set |
| Graph | How do supported features connect? | Coordinates, strand and splice edges | A locus-specific splice graph |
| Event | Which alternate paths share boundaries? | Graph traversal and event rules | Annotated or predicted splice events |
| Quantify | How frequently is each path used? | Boundary-spanning compatible reads | Relative splice-variant frequencies |
Inside this guide
1–12 · Foundations and design
- 1. Begin with genome-guided evidence
- 2. Separate prediction from quantification
- 3. Prepare splice-aware alignments
- 4. Use annotation without becoming trapped by it
- 5. Predict splice junctions
- 6. Predict exons from coverage
- 7. Store features with stable identities
- 8. Name independent samples
- 9. Combine samples deliberately
- 10. Build a splice graph
- 11. Respect strand direction
- 12. Define shared event boundaries
13–24 · Models and estimation
- 13. Recognise simple events
- 14. Represent complex events
- 15. Distinguish local paths from isoforms
- 16. Quantify compatible reads
- 17. Calculate relative frequencies
- 18. Keep low coverage visible
- 19. Use sample-level matrices
- 20. Visualise the splice graph
- 21. Interpret unannotated events cautiously
- 22. Inspect mapping plausibility
- 23. Inspect coverage uniformity
- 24. Test prediction thresholds
25–36 · Diagnostics and interpretation
- 25. Test annotation sensitivity
- 26. Test sample influence
- 27. Compare with SplAdder fairly
- 28. Compare with MAJIQ fairly
- 29. Connect to network thinking
- 30. Connect to fractions
- 31. Connect to computing
- 32. Connect to education pathways
- 33. Use bounded language
- 34. Ask the parent question
- 35. Did you know?
- 36. Finish with a reproducible bundle
Section 1 of 36
1. Begin with genome-guided evidence
SGSeq works from RNA-seq reads aligned to a reference genome. Genome coordinates make splice boundaries explicit, while the observed reads determine which exon and junction features enter the analysis.
Working checkpoint — transcript feature: Archive genome build, aligner version and full command. 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 prediction from quantification
Feature and event prediction decide what structures are available to measure; quantification estimates usage of those structures. Changing the catalogue can change the denominator even when read files stay identical.
Working checkpoint — splice graph: Version the feature set before comparing samples. 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. Prepare splice-aware alignments
Reads that bridge introns provide junction evidence, while contiguous coverage supports exon-like regions. Overhang thresholds, multimapping and strand information affect the graph that can be inferred.
Working checkpoint — splice event: Inspect several known loci before project-wide prediction. 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 annotation without becoming trapped by it
Existing transcript models can seed or annotate features, but the primary method is designed to predict unannotated and complex splice events from aligned RNA-seq. Annotation should guide interpretation without erasing observed alternatives.
Working checkpoint — event boundary: Record whether each feature was annotated, predicted or both. 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. Predict splice junctions
A splice junction connects a donor and acceptor coordinate supported by split reads. Junction evidence is local: it establishes a connection but not every exon carried by the full molecule.
Working checkpoint — variant frequency: Report junction coordinates, strand and supporting counts. 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. Predict exons from coverage
SGSeq derives exon-like transcript features from aligned reads and splice boundaries. Coverage breaks can reflect biology, bias or insufficient depth, so predicted exons need quality review.
Working checkpoint — genome-guided prediction: Compare predicted boundaries with coverage and annotation. 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. Store features with stable identities
The package represents transcript features with genomic coordinates, types and groupings. Stable identities allow quantification, plotting and comparison to refer to the same graph.
Working checkpoint — transcript feature: Check uniqueness and chromosome naming before merging objects. 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. Name independent samples
Biological replicates support group-level inference; multiple sequencing lanes or repeated BAM processing do not create new organisms or specimens.
Working checkpoint — splice graph: Build a sample table that distinguishes biological and technical units. 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. Combine samples deliberately
Pooling evidence across samples can reveal weak features, but a feature discovered mostly in one group may receive a different detection opportunity from a feature shared across groups.
Working checkpoint — splice event: Document the discovery set and consider a common catalogue. 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. Build a splice graph
Exons or splice boundaries form nodes and splice connections form edges. The graph summarises local routes supported or annotated at a locus; it is not a list of guaranteed full-length transcripts.
Working checkpoint — event boundary: Draw one graph with evidence counts on every edge. 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. Respect strand direction
Donor and acceptor meanings depend on transcript strand. A coordinate pair viewed in genomic order can reverse its biological direction on the minus strand.
Working checkpoint — variant frequency: Audit representative events on both strands. 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. Define shared event boundaries
In the SGSeq framework, a splice event is delimited by a start and end node connected by alternative paths. Reads spanning the event boundaries help isolate evidence for the competing variants.
Working checkpoint — genome-guided prediction: State the two boundaries before naming the event type. 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. Recognise simple events
Exon skipping and alternative splice-site choices can appear as compact graph alternatives. Their apparent simplicity still depends on the discovered features and the reads compatible with each path.
Working checkpoint — transcript feature: Keep coordinates and path definitions beside the familiar label. 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. Represent complex events
Nested or multi-path structures can be represented recursively rather than forced into one catalogue of only simple patterns. Complexity is useful when the graph supports it, but it also demands clearer visualisation.
Working checkpoint — splice graph: Show every alternative path contributing to the event. 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. Distinguish local paths from isoforms
Two paths through one event describe local splice variants. Short reads at the boundaries generally cannot phase that choice with distant exons elsewhere in the gene.
Working checkpoint — splice event: Avoid naming a complete isoform unless full-length evidence supports it. 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. Quantify compatible reads
SGSeq assigns evidence to event variants using reads compatible with the alternative paths and event boundaries. Ambiguous reads should not be treated as uniquely informative.
Working checkpoint — event boundary: Report informative and ambiguous evidence separately when possible. 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. Calculate relative frequencies
Variant frequencies describe how event evidence is distributed among alternatives. The value is conditional on the event definition and detectable reads rather than a universal transcript percentage.
Working checkpoint — variant frequency: Show counts and denominator beside the frequency. 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. Keep low coverage visible
An extreme frequency with very few informative reads is fragile. Missing or weak evidence should remain distinguishable from a well-supported zero.
Working checkpoint — genome-guided prediction: Set and report coverage criteria before interpretation. 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. Use sample-level matrices
Event-by-sample quantities support comparison and visualisation across the study. Sample order, identifiers and library metadata must remain aligned with the measurement matrix.
Working checkpoint — transcript feature: Assert exact sample-name matches in code. 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. Visualise the splice graph
The package provides graph and event plotting functions that place features, alternative paths and quantitative information together. A useful plot should let readers trace the numeric denominator.
Working checkpoint — splice graph: Export static plots for every headline event. 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. Interpret unannotated events cautiously
An unannotated event is absent from the chosen transcript catalogue, not necessarily absent from biology until now. Reference release and prediction filters define that label.
Working checkpoint — splice event: Say “unannotated in release X” rather than “new”. 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. Inspect mapping plausibility
Repetitive sequence, paralogues and short anchors can produce misleading junctions. A predicted graph inherits every alignment ambiguity that passed into the BAM files.
Working checkpoint — event boundary: Review decisive reads and mapping qualities in a genome browser. 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. Inspect coverage uniformity
RNA degradation, sequence composition and library preparation can create uneven exon coverage that affects feature prediction. Group-specific quality differences are especially dangerous.
Working checkpoint — variant frequency: Compare coverage and insert-size summaries across groups. 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. Test prediction thresholds
Feature-support thresholds control sensitivity and graph complexity. A permissive catalogue may admit noise; a strict one may miss rare alternatives.
Working checkpoint — genome-guided prediction: Track which headline paths survive defensible thresholds. 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. Test annotation sensitivity
A second justified annotation can relabel predicted features or change transcript context without altering the underlying reads. Robust interpretation follows coordinates and evidence across releases.
Working checkpoint — transcript feature: Create a coordinate-based crosswalk. 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. Test sample influence
A graph edge observed in one specimen may disappear when that specimen is removed. Discovery robustness and usage robustness should both be checked.
Working checkpoint — splice graph: Repeat key predictions without each independent sample. 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. Compare with SplAdder fairly
Both tools use splicing graphs, but their augmentation, event extraction and quantification rules differ. One complex locus may yield different event decompositions without either output being a simple error.
Working checkpoint — splice event: Translate outputs back to local nodes and edges. 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. Compare with MAJIQ fairly
MAJIQ reports local splicing variations with probabilistic evidence, while SGSeq predicts bounded graph events and relative variant frequencies. Their units and uncertainty summaries are not interchangeable.
Working checkpoint — event boundary: Align boundaries before comparing a result. 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. Connect to network thinking
A splice graph resembles a transport network with shared stations and alternative routes. The analogy helps students reason about paths while the read evidence keeps the biology grounded.
Working checkpoint — variant frequency: Ask which edge is observed and which route remains inferred. 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. Connect to fractions
Relative variant frequency is a fraction whose meaning depends on the competing paths in its denominator. Adding a newly predicted path can change every share.
Working checkpoint — genome-guided prediction: Recalculate a classroom fraction after adding one alternative. 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. Connect to computing
Graph traversal, genomic intervals and sparse sample matrices show how algorithms become part of measurement. Reproducible science must preserve both code and references.
Working checkpoint — transcript feature: Save package versions and checksums with outputs. 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. Connect to education pathways
Splice-event analysis links biology, mathematics and computer science and can illuminate routes into bioinformatics or biotechnology without locking a student into one future.
Working checkpoint — splice graph: Use official programme pages for current entry requirements. 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. Use bounded language
An SGSeq result supports a predicted local splice event and quantified alternative paths under the stated feature catalogue. It does not alone prove protein consequences or a causal mechanism.
Working checkpoint — splice event: Name the validation needed for the next claim. 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. Ask the parent question
A helpful question is: what reads cross the event boundaries, and are the same alternatives visible across independent samples?
Working checkpoint — event boundary: That answer reveals more than a software popularity ranking. 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?
SGSeq can predict nested splice events recursively, allowing a complex graph to be represented as related local choices rather than one oversized transcript puzzle.
Working checkpoint — variant frequency: Complexity should increase explanation, not confidence. 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 reference files, alignment records, sample metadata, SGSeq version, predicted features, splice graphs, events, sample matrices, plots, thresholds and complete sensitivity results.
Working checkpoint — genome-guided prediction: A prediction becomes useful when its construction and quantification can both be repeated. 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