eduKateSG · Why Science?
Run every allele through a versioned genome-feature map—and read impact labels as triage, not verdicts
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. SnpEff annotates genetic variants by comparing them with a versioned genome-feature database and writes predicted effects into structured fields such as ANN. It can report affected genes and transcripts, Sequence Ontology-style effects, coding changes and broad impact categories. The official documentation also provides summary outputs and database-building routes. A HIGH, MODERATE, LOW or MODIFIER category is a sorting aid under the supplied annotation—not proof of pathogenicity, mechanism or clinical importance. 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: Ensembl VEP and transcript-aware variant consequences; SQUIRLS and splice-variant prioritisation; SpliceAI and sequence-based splice-variant prediction; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official SnpEff and SnpSift documentation; Official SnpEff command-line documentation; Official SnpEff paper and citation page; SnpEff 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 | Which genome and feature models define the search space? | Assembly-matched SnpEff database and configuration | A versioned annotation environment |
| Annotate | Which features overlap each allele? | Normalised VCF and transcript structures | ANN records for affected alleles and transcripts |
| Prioritise | Which predicted effects deserve earlier review? | Effect terms and impact categories | A triage order, not a medical conclusion |
| Validate | Does measured evidence agree? | RNA, protein, population and experimental evidence | A defensible claim limited by validation |
Inside this guide
1–12 · Foundations and design
- 1. Start with a genome database
- 2. Match the assembly exactly
- 3. Understand interval-based annotation
- 4. Prepare a clean VCF
- 5. Preserve allele identity
- 6. Read the ANN field structure
- 7. Read effect terms before impact
- 8. Treat HIGH as a review flag
- 9. Treat MODERATE as context-dependent
- 10. Treat LOW and MODIFIER carefully
- 11. Retain all transcript annotations
- 12. Check transcript biotypes
13–24 · Models and estimation
- 13. Review amino-acid changes
- 14. Review splice-region calls
- 15. Review upstream and downstream windows
- 16. Build custom databases cautiously
- 17. Check coding-sequence consistency
- 18. Use summary statistics diagnostically
- 19. Separate SnpEff from SnpSift
- 20. Keep filters reviewable
- 21. Compare with Ensembl VEP fairly
- 22. Test indels and compound alleles
- 23. Inspect complex transcripts
- 24. Track database updates
25–36 · Diagnostics and interpretation
- 25. Audit one ANN record by hand
- 26. Inspect missing annotations
- 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 the result responsibly
- 35. Build a validation ladder
- 36. Finish with a reproducible bundle
Section 1 of 36
1. Start with a genome database
SnpEff needs a database that connects reference coordinates to genes, transcripts, exons and coding sequences. The database is part of the model, not background decoration.
Working checkpoint — genome database: Record the database name, build command or download source and checksum. 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. Match the assembly exactly
A VCF from one assembly cannot safely be annotated against features from another. Coordinate agreement is a precondition for every later effect term.
Working checkpoint — ANN field: Verify contig names, lengths and reference alleles before annotation. 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. Understand interval-based annotation
SnpEff uses genomic feature intervals and transcript structures to determine which variants overlap which biological regions. The resulting label depends on those boundaries.
Working checkpoint — effect term: Inspect one exon-boundary case against the source annotation. 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. Prepare a clean VCF
Split multiallelic records and normalise indels where appropriate. Equivalent alleles represented differently can otherwise produce inconsistent or duplicated results.
Working checkpoint — impact category: Keep the untouched callset and a transformation log. 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. Preserve allele identity
One VCF record may contain several alternative alleles. ANN subrecords must be matched back to the correct allele rather than attached by position alone.
Working checkpoint — transcript record: Retain REF and ALT beside every extracted annotation row. 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. Read the ANN field structure
The official command-line examples show pipe-delimited ANN values containing allele, effect, impact, gene, feature, biotype and coding details. Field order is a schema.
Working checkpoint — orthogonal evidence: Parse by the documented schema, not by manual string counting. 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. Read effect terms before impact
The effect term describes the predicted relationship, while the impact category groups consequences broadly. A specific term often communicates more science than the coarse category.
Working checkpoint — genome database: Report both fields and explain which one drove prioritisation. 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. Treat HIGH as a review flag
A predicted frameshift, stop-gained or essential splice-site consequence may receive high impact. That category says the annotation could be disruptive if the transcript is relevant and the call is correct.
Working checkpoint — ANN field: Inspect transcript relevance, allele quality and validation before escalation. 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. Treat MODERATE as context-dependent
Missense and other protein-altering predictions often sit in a moderate category. Their biological effect can range from negligible to important.
Working checkpoint — effect term: Add protein, conservation, frequency and functional evidence 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 10 of 36
10. Treat LOW and MODIFIER carefully
Synonymous, intronic, upstream or intergenic annotations can still matter in special regulatory contexts. Lower category does not mean impossible effect.
Working checkpoint — impact category: Avoid using impact alone as an exclusion rule without 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 36
11. Retain all transcript annotations
The same allele can be coding in one transcript, intronic in another and upstream of a third. Collapsing too early hides this transcript logic.
Working checkpoint — transcript record: Keep complete ANN records and document any transcript-selection rule. 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. Check transcript biotypes
Protein-coding, non-coding and other transcript biotypes support different consequence interpretations. A coding label requires an appropriate coding model.
Working checkpoint — orthogonal evidence: Include biotype and transcript version in review tables. 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. Review amino-acid changes
Protein-change notation follows the selected transcript and coding sequence. It predicts a sequence consequence but does not measure the protein in a sample.
Working checkpoint — genome database: Link the amino-acid change to its exact transcript and exon. 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. Review splice-region calls
Splice donor, acceptor and splice-region terms depend on configured positional definitions and annotation boundaries. The label is not RNA evidence.
Working checkpoint — ANN field: Use junction or targeted transcript measurements for confirmation. 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. Review upstream and downstream windows
Distance windows determine which nearby genes receive regulatory-location annotations. Changing the window changes the number of labels.
Working checkpoint — effect term: Record window settings and avoid causal language from proximity alone. Treat identifiers, denominators and factor levels as scientific quantities: a silent mismatch can reverse an otherwise correct calculation.
For parents, a calm checkpoint is to ask how many independent samples support the claim and which assumption matters most.
Section 16 of 36
16. Build custom databases cautiously
The documentation supports building databases for organisms or annotations beyond prebuilt sets. Input GFF, GTF or GenBank quality becomes part of output quality.
Working checkpoint — impact category: Validate feature counts, CDS phases and known test variants after building. 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. Check coding-sequence consistency
Transcript and protein sequences can reveal frame, start-codon or phase problems in a custom database. A database that builds is not necessarily biologically coherent.
Working checkpoint — transcript record: Run the available database checks and archive their warnings. 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. Use summary statistics diagnostically
SnpEff can create output summaries showing effect distributions and genomic regions. Unexpected proportions can expose build, filtering or annotation errors.
Working checkpoint — orthogonal evidence: Compare summaries with callset design before celebrating unusual biology. 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. Separate SnpEff from SnpSift
SnpEff annotates predicted effects, while SnpSift supports filtering and manipulation. A final shortlist may reflect both annotation and later selection rules.
Working checkpoint — genome database: Document every SnpSift expression separately from SnpEff annotation. 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. Keep filters reviewable
A chain of filters can silently remove variants with missing fields or unusual transcript records. The exclusion logic is part of the scientific conclusion.
Working checkpoint — ANN field: Report counts before and after each filter and preserve rejected 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 21 of 36
21. Compare with Ensembl VEP fairly
VEP and SnpEff can differ because of transcript catalogues, releases, normalisation and severity hierarchies. Raw agreement rates are uninterpretable without aligned references.
Working checkpoint — effect term: Run matched inputs and explain annotation-source differences. 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 indels and compound alleles
A local effect caller may examine one record at a time. Nearby variants on the same haplotype can alter the combined codon or reading frame.
Working checkpoint — impact category: Flag clustered alleles for haplotype-aware review. 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 complex transcripts
Overlapping genes, alternative starts, incomplete CDS records and retained introns can create many plausible annotations. More rows reflect model complexity, not software indecision.
Working checkpoint — transcript record: Show the transcript structures for high-stakes cases. 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. Track database updates
Prebuilt database names and external annotations evolve with releases. Re-running without a frozen version can change impact counts.
Working checkpoint — orthogonal evidence: Pin tool and database versions in containers or environment records. 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. Audit one ANN record by hand
Choose one variant and trace its genomic interval, transcript strand, coding position, codon change, amino-acid change, effect and impact.
Working checkpoint — genome database: Use the manual audit as a regression test for future updates. 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 missing annotations
Unplaced contigs, absent transcripts or malformed records may receive no useful ANN value. Missing output is not evidence of no effect.
Working checkpoint — ANN field: Retain unannotated counts and explicit failure reasons. 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
SnpEff predicts structural consequences from annotation. It does not integrate every criterion required for clinical classification.
Working checkpoint — effect term: Write “predicted high impact on transcript X” rather than “pathogenic”. 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
Sorting cards by observable features shows how classification rules work. Changing the rule can change the group without changing the card.
Working checkpoint — impact category: Use invented examples and avoid health conclusions. 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
DNA variants can affect coding sequence, RNA processing or nearby regions. SnpEff makes those possible relationships visible across a genome.
Working checkpoint — transcript record: Distinguish sequence position, predicted effect and measured phenotype. 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
Intervals, categories and frequencies make genome-wide summaries quantitative. Every percentage needs the filtered callset as its denominator.
Working checkpoint — orthogonal evidence: Label the denominator on each summary plot. 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
Efficient interval searches let millions of variants be compared with genome features. Schemas and parsers preserve meaning at scale.
Working checkpoint — genome database: Test ANN parsing with multiple alleles and multiple transcripts. 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
Genome annotation joins biology, data structures and software engineering and can illuminate STEM education and career pathways.
Working checkpoint — ANN field: Use official education sources for current programme choices. 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?
SnpEff can annotate many transcripts for one allele and place the structured results inside the VCF. The original variant and its annotations can therefore travel together.
Working checkpoint — effect term: Preserve the full VCF header that defines the ANN schema. 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 the result responsibly
A reader-ready table names assembly, database, transcript, effect, impact, coding change, filters and evidence status.
Working checkpoint — impact category: Do not let a colour-coded impact cell replace the biological explanation. 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
Check variants and reference, verify database features, compare annotation engines, inspect RNA or protein evidence and use experiments where mechanism matters.
Working checkpoint — transcript record: Validate the weakest step rather than repeating the same 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 36 of 36
36. Finish with a reproducible bundle
Deliver original and normalised VCFs, SnpEff version, config and database identity, command, full annotated VCF, summaries, filters, warnings and validation status.
Working checkpoint — orthogonal evidence: Make every shortlist row traceable to the complete ANN record. 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