eduKateSG · Why Science?
Place each variant into its transcript context, name every possible consequence, and keep annotation separate from proof
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. Ensembl Variant Effect Predictor, usually shortened to Ensembl VEP, accepts genomic variants and reports how each allele overlaps genes, transcripts, proteins and regulatory features. The current Ensembl documentation offers web, command-line and REST interfaces and can add reference information such as known-variant matches, population frequencies and predictor outputs. VEP is an annotation engine: it translates coordinates into possible consequences under a chosen release and configuration. A consequence label is not direct evidence that the allele changes health, molecular function or phenotype. 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: SpliceAI and sequence-based splice-variant prediction; MMSplice and modular splice-variant effects; RegTools and variant-aware splice-junction analysis; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official Ensembl Variant Effect Predictor documentation; Official Ensembl VEP command-line documentation; Official Ensembl VEP repository; Ensembl VEP primary Genome Biology study; 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 |
|---|---|---|---|
| Normalise | Which allele and coordinate are being annotated? | Genome build, reference FASTA and normalised variant | A reproducible input allele |
| Overlap | Which transcripts or regulatory features intersect it? | Release-matched Ensembl annotation | One or more feature-specific records |
| Describe | What consequence follows under that feature model? | Sequence Ontology terms and allele context | Transcript-, protein- or regulatory consequence labels |
| Validate | Does independent evidence support the biological claim? | Population, functional, segregation and experimental evidence | A bounded interpretation, not an automatic diagnosis |
Inside this guide
1–12 · Foundations and design
- 1. Begin with coordinates, not a gene story
- 2. Understand one-to-many annotation
- 3. Use Sequence Ontology terms
- 4. Choose the reference assembly
- 5. Normalise variant representation
- 6. Check reference alleles
- 7. Choose an interface deliberately
- 8. Freeze the Ensembl release
- 9. Prefer a local cache for reproducibility
- 10. Name the transcript source
- 11. Read protein consequences carefully
- 12. Read splice consequences carefully
13–24 · Models and estimation
- 13. Read regulatory consequences carefully
- 14. Distinguish consequence from severity
- 15. Understand most severe consequence
- 16. Use canonical flags as filters, not truth
- 17. Add population frequencies responsibly
- 18. Add predictor plugins transparently
- 19. Avoid score soup
- 20. Audit allele-specific matching
- 21. Test insertion and deletion edge cases
- 22. Test structural variants separately
- 23. Inspect mitochondrial conventions
- 24. Compare releases as a sensitivity test
25–36 · Diagnostics and interpretation
- 25. Validate HGVS output
- 26. Inspect warnings and missing fields
- 27. Use bounded clinical language
- 28. Connect to Primary Science
- 29. Connect to Secondary Science
- 30. Connect to mathematics
- 31. Connect to computing
- 32. Connect to pathways
- 33. Did you know?
- 34. Write a reader-ready result
- 35. Build a validation ladder
- 36. Finish with a reproducible bundle
Section 1 of 36
1. Begin with coordinates, not a gene story
VEP starts from a variant location and allele change. Gene names arrive only after that input is compared with a defined annotation release.
Working checkpoint — transcript consequence: Record chromosome, position, reference allele, alternative allele and genome assembly before interpretation. 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. Understand one-to-many annotation
One genomic allele can overlap several transcripts, each with different exon boundaries, coding frames or biotypes. The scientifically honest output is often several consequence rows.
Working checkpoint — Sequence Ontology term: Retain every relevant transcript row until the reporting rule has been declared. 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. Use Sequence Ontology terms
VEP reports standard consequence terms such as missense_variant, stop_gained or splice_donor_variant. A shared vocabulary makes pipelines comparable while preserving the need for biological context.
Working checkpoint — most severe consequence: Store the exact term rather than replacing it with an informal 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 4 of 36
4. Choose the reference assembly
GRCh37, GRCh38 and other assemblies assign coordinates against different reference sequences. A position without its assembly is incomplete scientific information.
Working checkpoint — allele normalisation: Make the pipeline stop when the VCF header, FASTA and cache disagree. 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. Normalise variant representation
Equivalent insertions or deletions can be written at different positions in repetitive sequence. Splitting multiallelic records and normalising alleles reduces accidental duplicate or missed annotations.
Working checkpoint — cache release: Archive the original and normalised VCF plus the normalisation command. 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. Check reference alleles
A reference mismatch can signal the wrong assembly, an incorrectly represented indel or a contig-naming error. Continuing silently turns annotation into fiction.
Working checkpoint — independent validation: Audit mismatch counts and inspect every unexpected reference failure. 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. Choose an interface deliberately
The official documentation distinguishes a web interface for smaller interactive jobs, a configurable command-line tool for larger workflows and a REST interface for programmatic queries.
Working checkpoint — transcript consequence: Match the interface to scale, reproducibility and data-governance needs. 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. Freeze the Ensembl release
Transcripts, consequence definitions and linked databases evolve. Ensembl release 116 was current on the official page checked in October 2026, but an analysis must cite the release actually used.
Working checkpoint — Sequence Ontology term: Record VEP version, cache release, assembly and annotation source together. Preserve the complete tested family and every excluded feature; a polished shortlist is not a substitute for provenance.
For scientific writing, prefer “supports”, “is associated with” and “is predicted to” when stronger causal verbs are not justified.
Section 9 of 36
9. Prefer a local cache for reproducibility
The command-line documentation supports cache and offline modes. A versioned local cache reduces dependence on changing remote database state.
Working checkpoint — most severe consequence: Checksum cache files and keep the installer or container recipe. 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. Name the transcript source
Ensembl and other transcript sets may not contain identical models. Transcript choice can change coding status, exon number and predicted amino-acid consequence.
Working checkpoint — allele normalisation: State the transcript catalogue and any canonical or MANE-style selection rule. 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. Read protein consequences carefully
For coding transcripts, VEP can describe amino-acid substitutions, frameshifts or gained stops. Those labels follow the transcript model; they do not measure protein abundance or activity.
Working checkpoint — cache release: Separate a predicted coding consequence from protein-level evidence. Keep the independent sample, design, annotation and software version beside this decision so the evidence can be reconstructed.
For a classroom model, use invented counts and clearly label the activity as an analogy rather than a sequencing experiment.
Section 12 of 36
12. Read splice consequences carefully
A splice_donor_variant or splice_acceptor_variant label describes positional overlap with a defined splice region. It does not show that RNA splicing changed in the sample.
Working checkpoint — independent validation: Add RNA evidence or a validated splice predictor before strengthening the claim. 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. Read regulatory consequences carefully
Variants may overlap regulatory regions or transcription-factor binding features when those annotations are enabled. Overlap is a candidate mechanism, not a measured expression effect.
Working checkpoint — transcript consequence: Name the regulatory dataset and tissue context behind the feature. 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. Distinguish consequence from severity
A consequence term describes a predicted structural relationship. Severity rankings help sort records but cannot replace disease-, gene- or tissue-specific evidence.
Working checkpoint — Sequence Ontology term: Keep the complete set of terms beside any most-severe 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 36
15. Understand most severe consequence
Collapsing many transcript consequences to one headline is convenient for triage. It can also hide a benign consequence on one transcript and a severe label on another.
Working checkpoint — most severe consequence: Document the selection hierarchy and preserve unsummarised rows. 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. Use canonical flags as filters, not truth
Canonical or principal-transcript flags are valuable routing aids. Biology can still depend on an alternative isoform expressed in the relevant tissue.
Working checkpoint — allele normalisation: Treat transcript flags as metadata and review tissue-specific evidence separately. 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. Add population frequencies responsibly
VEP can report allele frequencies from linked resources. Frequency informs rarity under the represented populations and release; it does not alone prove benignity or pathogenicity.
Working checkpoint — cache release: Record the dataset, population field, allele match and version. 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. Add predictor plugins transparently
The current documentation lists access to tools such as AlphaMissense, CADD, SIFT, REVEL and SpliceAI through options or plugins. Each score has its own target and calibration.
Working checkpoint — independent validation: Name every enabled plugin, version, field and missing-value rule. 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. Avoid score soup
Combining many predictor columns can look comprehensive while counting correlated models repeatedly. More columns do not automatically mean more independent evidence.
Working checkpoint — transcript consequence: Group scores by evidence type and predefine how disagreement is handled. 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. Audit allele-specific matching
External resources may store the same position with several alternative alleles. A correct annotation must match the allele, not just the coordinate.
Working checkpoint — Sequence Ontology term: Test multiallelic examples and retain the matched allele in every output. 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. Test insertion and deletion edge cases
Indels can cross exon boundaries, alter frames or have equivalent representations. Their consequence depends on normalisation, transcript orientation and exact sequence reconstruction.
Working checkpoint — most severe consequence: Hand-check a frameshift, in-frame indel and boundary-spanning case. 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. Test structural variants separately
VEP can annotate larger variant classes, but breakends, copy-number changes and incomplete coordinates require different assumptions from a single-nucleotide variant.
Working checkpoint — allele normalisation: Declare supported structural-variant representations and unresolved cases. 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 mitochondrial conventions
Mitochondrial contig names, genetic code and transcript resources differ from nuclear analysis. Generic defaults can mislabel an otherwise valid allele.
Working checkpoint — cache release: Use reference and annotation settings appropriate to the mitochondrial genome. 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. Compare releases as a sensitivity test
A new release may add, retire or redefine transcripts. Comparing a small critical set across justified releases reveals whether the headline depends on annotation drift.
Working checkpoint — independent validation: Report changed consequences instead of silently adopting the preferred one. 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. Validate HGVS output
HGVS strings are useful communication identifiers but depend on transcript version, sequence validation and formatting conventions. A plausible string can still be wrong.
Working checkpoint — transcript consequence: Retain the underlying genomic allele and transcript accession with version. 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. Inspect warnings and missing fields
Warnings about reference mismatch, incomplete transcripts or failed lookups are evidence about pipeline limits. Converting them to blank cells hides risk.
Working checkpoint — Sequence Ontology term: Keep warning codes and explicit missingness reasons in the final table. 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. Use bounded clinical language
VEP predicts consequences and collects annotations; it does not perform a complete clinical classification. Medical decisions require qualified interpretation under applicable standards.
Working checkpoint — most severe consequence: Write “annotated as” or “predicted to” unless stronger evidence exists. 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. Connect to Primary Science
A classification table can show how the same object receives different labels under different rules. This is a safe analogy for transcript-specific annotation.
Working checkpoint — allele normalisation: Use invented coloured cards rather than real patient variants. 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 Secondary Science
Genes are transcribed into RNA, and transcripts can be processed differently. VEP makes that information chain visible while showing that sequence context matters.
Working checkpoint — cache release: Mark which statements are coordinates, annotations, predictions or observations. 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 mathematics
One input can map to a set of outputs. Ranking and filtering then reduce that set, so denominators and decision rules must remain visible.
Working checkpoint — independent validation: Count transcript rows before and after each filter. 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
VCF, FASTA, cache and annotation schemas must agree. Version pinning and assertions are part of the scientific apparatus.
Working checkpoint — transcript consequence: Build a regression set containing substitutions, indels, multiallelic sites and splice boundaries. 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 pathways
Variant annotation brings together genetics, databases, statistics and software engineering and can illuminate STEM education and career pathways.
Working checkpoint — Sequence Ontology term: Check current official course information before making education decisions. 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. Did you know?
The official Ensembl VEP page offers web, command-line and REST routes to the same broad task. The interface changes, but the need to freeze reference data and options does not.
Working checkpoint — most severe consequence: Choose the route that leaves the clearest reproducible receipt. 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. Write a reader-ready result
A responsible result names the allele, assembly, transcript accession, consequence terms, selected summary rule, data release and validation status.
Working checkpoint — allele normalisation: Put the narrowest supported conclusion in the headline. 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. Build a validation ladder
Start with coordinate and allele checks, inspect transcript structure, compare justified annotations, add population and functional evidence, then validate the biological effect.
Working checkpoint — cache release: Do not skip from consequence label to diagnosis. 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 original and normalised VCFs, FASTA identity, VEP and cache releases, command or API request, plugins, full output, warnings, filters and validation record.
Working checkpoint — independent validation: Make every reported consequence traceable to an unsummarised feature-specific row. 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