eduKateSG · Why Science?
Build local splice events from observed junctions, then let missing evidence stay missing
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. Outrigger is a program for detecting and quantifying alternative splicing from junction reads. It can build a de novo event index, optionally validate splice sites and calculate Percent Spliced In for local skipped-exon and mutually exclusive-exon structures. It was developed in the context of sparse single-cell RNA-seq, where disciplined missingness and the distinction between cells and independent biological samples are essential. 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: LeafCutter and annotation-free intron excision; MAJIQ and local splicing variations; 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 Outrigger documentation; Outrigger index documentation; Outrigger PSI documentation; Expedition and Outrigger primary study record in PubMed; Expedition and Outrigger primary study full text; 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 |
|---|---|---|---|
| Index | Which local alternative exons are supported? | STAR junction files or alignments plus optional annotation | A de novo SE and MXE event database |
| Validate | Do introns have allowed splice-site sequences? | Reference genome and motif rules | A filtered structural catalogue |
| Quantify | Is read support sufficient for a PSI value? | Event junction counts and minimum-read rules | PSI or an explicit missing/uneventful state |
| Interpret | Is the pattern biological and reproducible? | Cell, donor, QC and orthogonal evidence | A bounded single-cell splicing hypothesis |
Inside this guide
1–12 · Foundations and design
- 1. Start with junction-spanning reads
- 2. Understand de novo event indexing
- 3. Prepare STAR junction outputs
- 4. Keep coordinate systems aligned
- 5. Build skipped-exon events
- 6. Build mutually exclusive-exon events
- 7. Recognise the supported scope
- 8. Use annotation as a supplement
- 9. Validate splice-site motifs
- 10. Keep non-canonical possibilities visible
- 11. Calculate PSI from event junctions
- 12. Respect minimum-read thresholds
13–24 · Models and estimation
- 13. Let missing stay missing
- 14. Distinguish zero from no evidence
- 15. Name the biological unit
- 16. Account for cell quality
- 17. Avoid coverage-driven cell types
- 18. Summarise cells without pseudoreplication
- 19. Inspect event support per cell
- 20. Inspect aggregate junction support
- 21. Check strandedness
- 22. Check multimapping and paralogues
- 23. Check one influential donor
- 24. Check event-index sensitivity
25–36 · Diagnostics and interpretation
- 25. Check annotation sensitivity
- 26. Check aligner sensitivity
- 27. Interpret trajectories cautiously
- 28. Separate modality from mean PSI
- 29. Compare with LeafCutter fairly
- 30. Compare with Whippet fairly
- 31. Connect to missing-data literacy
- 32. Connect to graph literacy
- 33. Connect to fair testing
- 34. Connect to education pathways
- 35. Use bounded claim language
- 36. Build a validation ladder
37–38 · Validation, learning and pathways
Section 1 of 38
1. Start with junction-spanning reads
Outrigger uses reads crossing exon–exon boundaries to identify local splice connections. Those reads provide direct support for a junction, but not complete full-length transcript structure.
Working checkpoint — junction read: Draw the genomic interval skipped by each junction. 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 38
2. Understand de novo event indexing
The index step discovers alternative exons and event combinations from observed junctions, supplementing exon definitions with annotation when supplied. De novo means data-derived within this reference and dataset.
Working checkpoint — de novo index: Avoid using novel as a synonym for functional. 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 38
3. Prepare STAR junction outputs
The documentation supports STAR junction files and alignment-derived inputs. Genome build, STAR parameters, read overhang and filtering determine which junctions are available for indexing.
Working checkpoint — skipped exon: Archive STAR version, command and SJ.out.tab files. 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 38
4. Keep coordinate systems aligned
Junction files, genome FASTA and annotation must agree on chromosome names, strand and coordinates. A small mismatch can create false novelty or prevent exon discovery.
Working checkpoint — mutually exclusive exon: Hand-check several well-known junctions. 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 38
5. Build skipped-exon events
A skipped-exon event contains junctions supporting inclusion through an alternative exon and a junction bypassing it. The event is local and does not identify the rest of the RNA molecule.
Working checkpoint — percent spliced in: Report the alternative and flanking exon coordinates. 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 38
6. Build mutually exclusive-exon events
An MXE event represents alternative internal exons connected to common flanking structure. Junction evidence supports local alternatives but short reads may not prove absolute exclusivity across every transcript.
Working checkpoint — missingness: Use long-read or targeted evidence for strict phasing claims. 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 38
7. Recognise the supported scope
The documented index focuses on SE and MXE event types. Other splicing patterns can exist in the same data but are outside this catalogue rather than biologically absent.
Working checkpoint — junction read: State the event classes included in every summary. 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 38
8. Use annotation as a supplement
Known exon coordinates can help define structures, while junction evidence can reveal alternatives not already represented. Annotation aids interpretation without replacing the observed read support.
Working checkpoint — de novo index: Freeze and cite the annotation release. 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 38
9. Validate splice-site motifs
The optional validate step can retain events whose introns use allowed splice-site sequences such as canonical motifs. Sequence plausibility is a useful filter, not proof of regulated biology.
Working checkpoint — skipped exon: Report allowed motifs and the genome used for lookup. 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 38
10. Keep non-canonical possibilities visible
Real non-canonical splicing exists, but sequencing and alignment artefacts also create unusual boundaries. Excluding or retaining these events should be a declared scientific choice.
Working checkpoint — mutually exclusive exon: Preserve a table of events removed by motif validation. 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 38
11. Calculate PSI from event junctions
The psi step uses junction counts for the indexed events to estimate the share of the included alternative. Each value belongs to a particular event definition and read-support rule.
Working checkpoint — percent spliced in: Show the junction components of the denominator. 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 38
12. Respect minimum-read thresholds
Outrigger requires sufficient evidence across event junctions before assigning a PSI value. This guards against extreme estimates from one or two reads.
Working checkpoint — missingness: Record threshold values and how many cell–event pairs pass. 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 38
13. Let missing stay missing
Sparse single-cell data often provide too little evidence for a valid PSI. A missing value can be the scientifically correct output; replacing it with zero falsely asserts exon exclusion.
Working checkpoint — junction read: Keep missing, zero and uneventful states distinct. 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 38
14. Distinguish zero from no evidence
A PSI of zero requires support for the exclusion path under the method’s rules. No qualifying reads means the event was not measured in that cell, not that inclusion was absent.
Working checkpoint — de novo index: Use explicit reason codes in downstream tables. 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 38
15. Name the biological unit
A cell is a measured unit, but cells from the same donor or culture are not automatically independent biological replicates. Group inference must respect the level of sampling or treatment assignment.
Working checkpoint — skipped exon: Report both cell count and donor count. 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 38
16. Account for cell quality
Library size, detected genes, mitochondrial proportion and alignment quality can correlate with the number of quantifiable events. Cell QC therefore shapes the observed splicing landscape.
Working checkpoint — mutually exclusive exon: Plot PSI availability against QC metrics. 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 38
17. Avoid coverage-driven cell types
A cell type with deeper libraries may appear to have more splicing complexity simply because more events pass thresholds. Detection opportunity must be separated from biological diversity.
Working checkpoint — percent spliced in: Match or model coverage where appropriate. 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 38
18. Summarise cells without pseudoreplication
Thousands of cells do not turn a three-donor study into a thousand-donor study. Subject-aware summaries, resampling or models are needed for population-level claims.
Working checkpoint — missingness: Make donor the visible replication level. 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 38
19. Inspect event support per cell
A PSI matrix can hide the actual junction counts. Headline events should be traced back to inclusion, exclusion and total qualifying evidence in representative cells.
Working checkpoint — junction read: Display counts beside PSI distributions. 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 38
20. Inspect aggregate junction support
Pooling evidence across cells can improve event discovery, while cell-level quantification remains sparse. The discovery pool and inference unit should be described separately.
Working checkpoint — de novo index: State which cells contributed to the index. 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 38
21. Check strandedness
Wrong strand handling can join unrelated genes or reverse event interpretation. Library protocol, STAR output and annotation strand must be consistent.
Working checkpoint — skipped exon: Verify strand on known events before indexing. 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 38
22. Check multimapping and paralogues
Single-cell reads are often short and sparse, so junctions in repetitive or homologous regions deserve special caution. A graph edge can be precise in coordinates and still be ambiguously mapped.
Working checkpoint — mutually exclusive exon: Inspect uniqueness around decisive boundaries. 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 38
23. Check one influential donor
A cell-level pattern may vanish when one donor is removed. Donor-level leave-one-out analysis reveals whether a result generalises beyond one individual.
Working checkpoint — percent spliced in: Lower confidence when donor direction is inconsistent. 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 38
24. Check event-index sensitivity
Minimum junction evidence and the set of cells used for discovery affect which events enter the database. A changed index can alter both denominators and missingness.
Working checkpoint — missingness: Rebuild key analyses under one planned threshold sensitivity. 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 38
25. Check annotation sensitivity
A different annotation may supply alternative exon boundaries or gene labels while the underlying junctions stay similar. Robust claims should survive reasonable catalogue changes.
Working checkpoint — junction read: Track events by coordinates, not only IDs. 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 38
26. Check aligner sensitivity
Novel junction calls can depend on STAR overhang and filtering rules. A defensible alternative alignment helps identify fragile edges.
Working checkpoint — de novo index: Keep genome and cell set 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 38
27. Interpret trajectories cautiously
The primary Expedition study used single-cell splicing to examine neuronal differentiation. A smooth computational trajectory is an ordering model, not direct observation of one cell transforming into another.
Working checkpoint — skipped exon: Separate cross-sectional ordering from lineage proof. 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 38
28. Separate modality from mean PSI
A group can contain cells near zero and one even when its average is intermediate. Bimodality or other distributions carry information that a mean alone erases.
Working checkpoint — mutually exclusive exon: Plot the full measured-cell distribution and missing fraction. 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 38
29. Compare with LeafCutter fairly
LeafCutter clusters excised introns and tests relative usage across samples; Outrigger builds explicit SE and MXE event graphs and assigns PSI only when junction evidence passes rules.
Working checkpoint — percent spliced in: Map the local junction sets before comparing. 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 38
30. Compare with Whippet fairly
Whippet quantifies contiguous splice graphs and complex events, whereas Outrigger’s documented event catalogue is deliberately narrower. Coverage, graph definition and uncertainty handling differ.
Working checkpoint — missingness: Compare the same biological splice paths, not tool labels. 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 38
31. Connect to missing-data literacy
A blank measurement is not automatically a zero. This lesson travels from single-cell RNA-seq to school experiments, surveys and sensor readings.
Working checkpoint — junction read: Ask students what evidence would justify recording zero. 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 38
32. Connect to graph literacy
Outrigger turns junctions into a graph database of local events. Nodes and edges make the logic inspectable and connect molecular biology with computing.
Working checkpoint — de novo index: Build a small paper graph and label observed 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 33 of 38
33. Connect to fair testing
Consistent thresholds, independent donors and transparent exclusions are modern forms of the same fair-test habits learned in Science.
Working checkpoint — skipped exon: Have students identify the comparison and control before interpreting PSI. 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 38
34. Connect to education pathways
Single-cell splicing joins biology, statistics and software engineering and can illuminate routes into life science and bioinformatics. It does not promise admission or a job.
Working checkpoint — mutually exclusive exon: Check current official programme information 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 38
35. Use bounded claim language
An Outrigger result supports a local SE or MXE proportion where sufficient junction evidence exists. It does not alone prove full isoforms, protein function or causation.
Working checkpoint — percent spliced in: State missingness and donor replication beside the effect. 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 38
36. Build a validation ladder
Junction review challenges mapping; targeted RNA assays challenge the event; long reads challenge phasing; perturbation and protein work challenge mechanism.
Working checkpoint — missingness: Choose the rung that can falsify the proposed story. 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 38
37. Did you know?
Outrigger can return no PSI when the junction evidence is insufficient, which protects readers from confusing an unmeasured event with complete exon exclusion.
Working checkpoint — junction read: Treat principled missingness as a result, not an inconvenience. 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 38
38. Finish with a reproducible bundle
Deliver genome, annotation, STAR settings, junction files, event index, validation rules, Outrigger version, cell and donor metadata, PSI outputs, reason codes, QC and sensitivity checks.
Working checkpoint — de novo index: A single-cell matrix is reproducible only when its missing values have provenance. 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.
Contents · Previous section · Continue to the Science Learning Hub