eduKateSG · Why Science?
Read allele frequency with its denominator, coverage, filters, ancestry group and release—because “rare” is never a complete conclusion
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. The Genome Aggregation Database, gnomAD, aggregates and harmonises human exome and genome sequencing data so researchers can inspect allele counts, frequencies, coverage and gene-constraint summaries. The official v4 release combined an exome callset from 730,947 individuals and a genome callset from 76,215 individuals, aligned to GRCh38; the 2026 flagship publication analyses the large v4 exome resource. These data provide population context, not a census of humanity and not a verdict about disease. Frequency must be read with the release, callset, denominator, quality filters, coverage and represented genetic-ancestry groups. 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: ClinVar and evidence-aware variant classifications; Ensembl VEP and transcript-aware variant consequences; ANNOVAR and gene-, region- and filter-based annotation; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official gnomAD browser; Official gnomAD v4 release notes; gnomAD v4 flagship study in PubMed; gnomAD population and constraint study 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 |
|---|---|---|---|
| Identify | Which allele, build, release and callset are being queried? | Normalised variant, GRCh38 context and gnomAD release | A reproducible lookup |
| Count | How many alternate and callable alleles were observed? | Allele count, allele number, filters and coverage | A denominator-aware frequency |
| Stratify | How does the observation vary across represented groups? | Genetic-ancestry group, sex and subpopulation summaries | A contextual comparison without stereotyping |
| Interpret | What does frequency support under the biological model? | Inheritance, penetrance, phenotype and independent evidence | A bounded conclusion, never causation from rarity alone |
Inside this guide
1–12 · Foundations and design
- 1. Start with a population reference
- 2. Do not call the cohort “healthy”
- 3. Freeze the release and assembly
- 4. Separate exomes from genomes
- 5. Read the v4 scale correctly
- 6. Know allele count and allele number
- 7. Inspect site filters
- 8. Inspect coverage
- 9. Treat absence as a bounded statement
- 10. Compare groups with denominators
- 11. Use genetic-ancestry labels carefully
- 12. Read group maximums cautiously
13–24 · Models and estimation
- 13. Handle sex chromosomes separately
- 14. Treat mitochondrial data as its own system
- 15. Treat structural variants as a separate callset
- 16. Understand rarity
- 17. Respect founder effects
- 18. Check relatedness and quality control
- 19. Inspect genotype quality
- 20. Normalise the allele
- 21. Handle multiallelic sites
- 22. Read constraint as gene-level evidence
- 23. Inspect predicted loss-of-function flags
- 24. Understand observed versus expected
25–36 · Diagnostics and interpretation
- 25. Watch for technical artefacts
- 26. Use frequency in a disease model
- 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. Start with a population reference
gnomAD pools sequence data from many contributing projects to create large summary resources. The result is a reference dataset, not a random sample of every person or place.
Working checkpoint — allele count: Name the release and represented callset before interpreting a frequency. 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. Do not call the cohort “healthy”
Contributing studies use different recruitment and inclusion rules. Population reference data can include people with adult-onset or unrecognised conditions.
Working checkpoint — allele number: Describe the dataset as aggregated sequencing data rather than a healthy control cohort. 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. Freeze the release and assembly
gnomAD releases differ in sample composition, reference build, calling and quality control. The official v4 callsets use GRCh38.
Working checkpoint — allele frequency: Record release, build, dataset and accessed date 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 4 of 36
4. Separate exomes from genomes
Exome sequencing targets coding regions, while genome sequencing covers a broader reference with different depth and calling properties. Counts from the two callsets are not interchangeable.
Working checkpoint — coverage: State which callset supplied every number. 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. Read the v4 scale correctly
The official v4 release notes describe 730,947 exome samples and 76,215 genome samples, for 807,162 total individuals across the two callsets.
Working checkpoint — genetic ancestry group: Cite the dated release rather than copying an undated rounded number. 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. Know allele count and allele number
Allele count is the observed alternate-allele total. Allele number is the number of called alleles contributing to that site; frequency is their ratio.
Working checkpoint — constraint metric: Report AC, AN and AF together. 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. Inspect site filters
A displayed allele may carry quality filters, and a missing site may have inadequate evidence for a reliable call. PASS status and filter details matter.
Working checkpoint — allele count: Do not use a frequency without recording filter status. 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. Inspect coverage
Low or uneven coverage reduces the callable denominator and can make absence uninformative. Exome capture designs especially vary at boundaries.
Working checkpoint — allele number: Open the coverage track for consequential absence claims. 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 absence as a bounded statement
No observed alternate allele in the displayed callset is not proof that the allele does not exist. Sampling, ancestry representation, coverage and filters limit the claim.
Working checkpoint — allele frequency: Write “not observed in this release and callset” with AN and coverage. 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. Compare groups with denominators
A group-specific frequency can look high or low when the allele number is small. Counts and confidence matter more than a decimal alone.
Working checkpoint — coverage: Show AC and AN for every compared group. 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. Use genetic-ancestry labels carefully
gnomAD uses genetic-ancestry group summaries to aid interpretation. These analytical labels are not interchangeable with nationality, culture, ethnicity or personal identity.
Working checkpoint — genetic ancestry group: Use the resource’s exact terminology and avoid essentialist 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 12 of 36
12. Read group maximums cautiously
A highest group frequency can reveal founder or sampling patterns, but it can also be unstable with small denominators.
Working checkpoint — constraint metric: Check counts, quality and subpopulation structure before applying a maximum. 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. Handle sex chromosomes separately
Ploidy, pseudoautosomal regions and sex-stratified callsets change denominators on X and Y. Autosomal assumptions can yield impossible frequencies.
Working checkpoint — allele count: Use chromosome-appropriate allele-number logic. 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. Treat mitochondrial data as its own system
Mitochondrial variants involve heteroplasmy, copy number and a separate release context. Nuclear-genome frequency rules do not transfer unchanged.
Working checkpoint — allele number: Use the dedicated mitochondrial fields and documentation. 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. Treat structural variants as a separate callset
Deletions, duplications and other structural variants use different discovery and QC methods from small variants.
Working checkpoint — allele frequency: Cite the structural-variant release and representation explicitly. 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. Understand rarity
A low allele frequency can be consistent with a rare disorder model, sequencing error, recent variation or incomplete sampling. Rarity is a clue, not a mechanism.
Working checkpoint — coverage: Combine frequency with inheritance, phenotype and functional 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 17 of 36
17. Respect founder effects
An allele can be uncommon globally but more frequent in one represented group because of demographic history. Global averages can hide that context.
Working checkpoint — genetic ancestry group: Review group-specific counts without turning group summaries into individual predictions. 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. Check relatedness and quality control
Large resources remove or account for related samples and apply sample and variant QC under release-specific methods. Those choices shape the summaries.
Working checkpoint — constraint metric: Read the methods for the release used in analysis. 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. Inspect genotype quality
Allele balance, depth and genotype likelihoods affect whether a variant is called. Aggregated counts inherit the strengths and limits of upstream calls.
Working checkpoint — allele count: Review quality histograms or flags for surprising results. 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. Normalise the allele
Insertions and deletions can be represented at equivalent nearby positions. A lookup may miss a real match when the query and database representations differ.
Working checkpoint — allele number: Normalise against the matching reference and verify REF and ALT. 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. Handle multiallelic sites
Several alternate alleles at one coordinate have distinct counts and consequences. Position-only matching transfers evidence incorrectly.
Working checkpoint — allele frequency: Match the exact alternate allele and split logic. 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. Read constraint as gene-level evidence
gnomAD compares observed and expected variation to estimate how depleted a gene is for classes such as predicted loss of function. Constraint summarises population variation, not one patient.
Working checkpoint — coverage: Keep variant-level frequency and gene-level constraint in separate columns. 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 predicted loss-of-function flags
Transcript choice and annotation quality can create false predicted loss-of-function calls. Curated flags and tools help, but do not replace manual review.
Working checkpoint — genetic ancestry group: Open the transcript and variant annotations behind a consequential pLoF count. 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. Understand observed versus expected
Constraint metrics depend on mutational models, callable regions and the selected transcript. The denominator and model define the comparison.
Working checkpoint — constraint metric: Record the metric version and transcript. 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. Watch for technical artefacts
Sequence context, mapping difficulty and platform effects can produce apparent variants or missing calls. Massive sample size does not erase systematic error.
Working checkpoint — allele count: Challenge outliers with read-level or orthogonal evidence when available. 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. Use frequency in a disease model
Plausible maximum frequency depends on prevalence, penetrance, inheritance and genetic heterogeneity. One universal rarity threshold is rarely defensible.
Working checkpoint — allele number: Write the disease-model assumptions beside the threshold. 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
gnomAD supports population context; it does not classify a variant or diagnose a person. Medical interpretation requires qualified, case-specific review.
Working checkpoint — allele frequency: Say “frequency is compatible or incompatible with the stated model” rather than “benign or 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
A jar-of-beads activity can show why seeing zero red beads in one scoop does not prove the jar contains none.
Working checkpoint — coverage: Keep the sample size visible in every comparison. 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
Alleles, inheritance and population variation become measurable through counts and denominators. Students can test how sampling changes apparent rarity.
Working checkpoint — genetic ancestry group: Separate biological diversity from social labels. 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
AF equals AC divided by AN, but confidence depends on the denominator. Stratification and missing calls change the sample space.
Working checkpoint — constraint metric: Recalculate frequencies from counts before comparing decimals. 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
Large exome and genome callsets require scalable storage, harmonised QC, indexed queries and versioned interfaces. Data engineering is part of the science.
Working checkpoint — allele count: Build regression tests for build, allele, filter and missingness handling. 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
Population genomics joins biology, statistics, ethics and distributed computation and can illuminate STEM education and career pathways.
Working checkpoint — allele number: Use current official programme information for real pathway 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?
A variant can be absent from hundreds of thousands of exomes yet still sit in a poorly covered interval. The denominator and coverage decide what “absent” can mean.
Working checkpoint — allele frequency: Read AN and coverage before celebrating a zero. 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 strong result names allele, build, release, callset, filters, coverage, AC, AN, AF, group context and the disease-model assumption.
Working checkpoint — coverage: Keep the counts beside every rounded frequency. 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
Verify allele identity, inspect quality and coverage, compare represented groups, evaluate the inheritance model, then add phenotype and functional evidence.
Working checkpoint — genetic ancestry group: Do not jump from rarity to causation. 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 the normalised query, gnomAD release and callset, screenshots or exported fields, API or browser route, all denominators, filters, coverage, model assumptions and review date.
Working checkpoint — constraint metric: Make every frequency statement reconstructable. 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