eduKateSG · Why Science?
Augment an annotation with read evidence, then reduce a complicated graph to testable events
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. SplAdder converts a transcript annotation into splicing graphs, augments those graphs with evidence from RNA-seq alignments, extracts alternative-splicing events and quantifies them for further analysis. Its current documentation separates build, test and visualisation modes. Graph confidence settings, read evidence, sample design and event definitions shape every conclusion. 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; LeafCutter and annotation-free intron excision; rMATS and replicate-aware differential splicing; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official SplAdder repository; Official SplAdder 3.0.2 documentation; SplAdder primary study in Bioinformatics; SplAdder 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 |
|---|---|---|---|
| Build | What annotated and observed connections belong in each locus? | GTF or GFF annotation, BAM alignments and confidence settings | An augmented splicing graph |
| Extract | Which graph patterns correspond to event classes? | Graph topology and event definitions | A versioned event catalogue |
| Quantify | How much evidence supports each alternative? | Junction counts and exon coverage | Event-level counts and PSI-like summaries |
| Test | Do event proportions differ across groups? | Independent samples, filters and a declared contrast | Differential event evidence |
Inside this guide
1–12 · Foundations and design
- 1. Start with the augmented graph
- 2. Prepare compatible annotation
- 3. Prepare consistent alignments
- 4. Understand graph vertices and edges
- 5. Augment with RNA-seq evidence
- 6. Choose a confidence level deliberately
- 7. Build a shared project
- 8. Name biological replicates
- 9. Extract exon-skipping events
- 10. Extract alternative splice sites
- 11. Extract intron-retention events
- 12. Extract mutually exclusive structures
13–24 · Models and estimation
- 13. Distinguish event catalogue from biology
- 14. Quantify junction evidence
- 15. Quantify exon coverage
- 16. Read PSI-like proportions
- 17. Filter sparse events
- 18. Use build mode reproducibly
- 19. Use test mode with a declared contrast
- 20. Use viz mode as evidence review
- 21. Correct for multiple testing
- 22. Pair evidence with effect size
- 23. Inspect sample heterogeneity
- 24. Inspect graph sensitivity
25–36 · Diagnostics and interpretation
- 25. Inspect annotation sensitivity
- 26. Inspect alignment sensitivity
- 27. Inspect one influential sample
- 28. Compare with rMATS fairly
- 29. Compare with LeafCutter fairly
- 30. Compare with MAJIQ fairly
- 31. Connect to network thinking
- 32. Connect to fair testing
- 33. Connect to computing
- 34. Connect to education pathways
- 35. Use bounded claim language
- 36. Did you know?
37–37 · Validation, learning and pathways
Section 1 of 37
1. Start with the augmented graph
SplAdder begins with annotated transcript structures and adds splice evidence found in RNA-seq alignments. The resulting graph can represent more diversity than the original annotation while retaining an explicit connection to it.
Working checkpoint — splicing graph: Draw annotated edges and added edges in different colours. 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 37
2. Prepare compatible annotation
The software accepts standard annotation formats, but coordinate, strand and chromosome conventions must match the alignments and genome. A formatting mismatch can silently remove whole loci.
Working checkpoint — graph augmentation: Validate several known genes before full-scale building. 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 37
3. Prepare consistent alignments
Every sample should use the same genome build and a documented splice-aware alignment policy. Differences in novel-junction discovery settings can look like biological differences later.
Working checkpoint — alternative event: Archive aligner versions, commands and QC. 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 37
4. Understand graph vertices and edges
Exonic segments become vertices and intronic connections become edges in a locus-specific splicing graph. The graph is a structured summary of possible paths, not proof that every complete path exists in one molecule.
Working checkpoint — junction evidence: Keep local connectivity separate from full-length phasing. 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 37
5. Augment with RNA-seq evidence
Observed reads can support new exons or introns that extend the initial annotation. Augmentation widens discovery, but weak mapping or contamination can also add misleading structure.
Working checkpoint — confidence level: Require transparent support for every headline addition. 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 37
6. Choose a confidence level deliberately
SplAdder provides graph-augmentation confidence settings that trade sensitivity against stringency. A permissive setting may recover rare biology and more noise; a strict setting may miss weak events.
Working checkpoint — differential testing: Record the setting and compare retained graph elements. 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 37
7. Build a shared project
The documented workflow operates on a project directory whose build products support later quantification, testing and visualisation. Reusing the wrong directory can mix incompatible references or samples.
Working checkpoint — splicing graph: Version the project directory and manifest. 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 37
8. Name biological replicates
Independent specimens or cultures support group inference. Multiple BAM files from technical lanes do not create new biological units unless the design genuinely sampled them independently.
Working checkpoint — graph augmentation: Resolve replicate identity before building group lists. 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 37
9. Extract exon-skipping events
A graph pattern can describe an exon included between flanking exons and an edge that bypasses it. This local event does not specify the rest of the transcript.
Working checkpoint — alternative event: Report all defining exon and junction coordinates. 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 37
10. Extract alternative splice sites
Alternative 5-prime or 3-prime events use competing donor or acceptor positions. Correct strand interpretation is essential because biological names reverse with transcript orientation.
Working checkpoint — junction evidence: Verify one event on each strand. 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 37
11. Extract intron-retention events
Retention contrasts an intronic segment present in RNA with the spliced connection that removes it. Pre-mRNA, DNA contamination and uneven coverage are important alternative explanations.
Working checkpoint — confidence level: Inspect intronic coverage and library preparation. 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 37
12. Extract mutually exclusive structures
Graph topology can identify alternatives corresponding to mutually exclusive exon choices. Short reads support local paths but may not prove exclusivity across complete transcripts.
Working checkpoint — differential testing: Use long reads for strict full-length 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 13 of 37
13. Distinguish event catalogue from biology
An extracted event is a computational pattern matching a definition. Several rows can describe overlapping aspects of one locus, and not every event implies an independent mechanism.
Working checkpoint — splicing graph: Group events by shared vertices and edges. 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 37
14. Quantify junction evidence
Splice-junction reads provide direct support for graph edges. Low overhang, multimapping and repetitive sequence can reduce their reliability even when counts are non-zero.
Working checkpoint — graph augmentation: Inspect decisive edges in the alignments. 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 37
15. Quantify exon coverage
Exon-body coverage complements junction evidence but can be affected by overlapping transcription, sequence bias and gene-expression changes. The evidence types should be kept visible.
Working checkpoint — alternative event: Report junction and exon components separately. 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 37
16. Read PSI-like proportions
Event inclusion is a relative quantity whose denominator includes competing alternatives. A rising proportion can result from more inclusion evidence, less exclusion evidence or both.
Working checkpoint — junction evidence: Show component counts beside the proportion. 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 37
17. Filter sparse events
Very weak events cannot support stable proportion estimates or group tests. Filtering defines the analysed family and therefore belongs in the methods and result counts.
Working checkpoint — confidence level: Publish every threshold and before–after total. 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 37
18. Use build mode reproducibly
The current documentation describes build mode for graph construction and event extraction. Parameters controlling sample discovery, annotation and graph augmentation must be saved with the outputs.
Working checkpoint — differential testing: Keep the exact command in the project bundle. 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 37
19. Use test mode with a declared contrast
Test mode compares specified sample groups using the project’s quantified events. File lists, labels and group order need the same care as a laboratory treatment sheet.
Working checkpoint — splicing graph: Print resolved sample paths before 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 20 of 37
20. Use viz mode as evidence review
Visualisation can show splice graphs and alternative events in genomic context. A persuasive picture should remain tied to numeric counts, settings and sample identities.
Working checkpoint — graph augmentation: Export static views 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 37
21. Correct for multiple testing
Many events may be compared. Adjusted evidence applies to the tested event family after filtering and does not convert a computational event into a causal mechanism.
Working checkpoint — alternative event: Retain full raw and adjusted tables. 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 37
22. Pair evidence with effect size
Statistical detectability and biological magnitude are different. A small change may be precise in a large study, while a large sparse change may remain unstable.
Working checkpoint — junction evidence: Report proportion change, counts and adjusted evidence together. 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 37
23. Inspect sample heterogeneity
Group summaries can conceal outliers or subgroups. Event-level sample plots show whether a change is shared across biological replicates.
Working checkpoint — confidence level: Place individual sample values over the group summary. 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 37
24. Inspect graph sensitivity
Changing confidence level can add or remove vertices and edges, which can alter event extraction as well as quantification. This is structural sensitivity, not merely a new p-value.
Working checkpoint — differential testing: Track event identities across planned settings. 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 37
25. Inspect annotation sensitivity
A newer transcript catalogue may absorb formerly novel edges or split loci differently. Robust conclusions should acknowledge whether the local pattern survives.
Working checkpoint — splicing graph: Rebuild key loci under one justified alternative annotation. 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 37
26. Inspect alignment sensitivity
Novel junctions can depend on aligner thresholds and overhang rules. A second defensible alignment provides evidence about measurement stability.
Working checkpoint — graph augmentation: Keep all other references fixed in the comparison. 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 37
27. Inspect one influential sample
Leave-one-replicate-out runs can expose an event driven by one specimen. This diagnostic should reduce confidence when conclusions are unstable.
Working checkpoint — alternative event: Report direction and event retention 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 28 of 37
28. Compare with rMATS fairly
rMATS detects standard event classes with replicate-aware modelling, while SplAdder explicitly augments graphs before extracting and testing events. The same label may arise from different catalogue construction.
Working checkpoint — junction evidence: Align coordinates and supporting paths first. 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 37
29. Compare with LeafCutter fairly
LeafCutter clusters observed introns without requiring a transcript annotation for discovery. SplAdder starts from annotation and augments a graph, so novelty and denominators differ.
Working checkpoint — confidence level: Compare the local splice connections, not total hit counts. 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 37
30. Compare with MAJIQ fairly
MAJIQ local splicing variations and SplAdder event classes are distinct graph reductions. Complex branches may become one LSV but several standard events.
Working checkpoint — differential testing: Translate each output back to vertices and edges. 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 37
31. Connect to network thinking
A splicing graph resembles a transport network in which alternate edges create routes. The analogy makes algorithms approachable while molecule-specific evidence keeps it scientifically grounded.
Working checkpoint — splicing graph: Ask which routes are observed locally and which remain inferred. 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 37
32. Connect to fair testing
A consistent build, independent repeats and a fixed contrast are the same fair-test habits taught in school science. Advanced code depends on those foundations.
Working checkpoint — graph augmentation: Have students identify variables and controls before discussing graphs. 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 37
33. Connect to computing
Graph traversal, interval arithmetic and project directories turn computational choices into part of the experiment. Reproducibility needs both code and reference data.
Working checkpoint — alternative event: Use checksums for annotations and build products. 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 37
34. Connect to education pathways
Splicing-graph analysis joins biology, mathematics and computer science and can illuminate study routes into bioinformatics and biotechnology. It is not an admission promise.
Working checkpoint — junction evidence: Check current official course requirements 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 35 of 37
35. Use bounded claim language
A SplAdder result supports an event structure and group difference under the stated graph, counts and test. It does not alone prove complete isoforms or functional consequences.
Working checkpoint — confidence level: Match the conclusion to the event-level evidence. 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 37
36. Did you know?
SplAdder’s name describes its central move: it adds RNA-seq-supported structure to an existing splice graph before extracting familiar alternative-splicing events.
Working checkpoint — differential testing: Treat every added edge as a hypothesis with evidence, not automatic truth. 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 37
37. Finish with a reproducible bundle
Deliver annotation, alignments or accessions, project manifest, confidence level, SplAdder version, commands, event catalogue, counts, complete tests, visualisations and sensitivities.
Working checkpoint — splicing graph: A graph can be trusted only when its construction can be repeated. 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.
Contents · Previous section · Continue to the Science Learning Hub