eduKateSG · Why Science?
Describe what was actually observed with precise, versioned phenotype terms before asking a computer—or a person—to connect the pattern
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 Human Phenotype Ontology, HPO, provides a standard vocabulary for phenotypic abnormalities encountered in human disease. Its terms have stable identifiers, definitions, synonyms and relationships that place specific observations beneath broader concepts. This makes clinical observations searchable and computable, including semantic-similarity analyses used in research and diagnostic support. HPO does not diagnose a person: the quality of any result depends on what was observed, what was explicitly absent, how specifically it was recorded, the individual’s age and context, the disease annotations available, and the release used. 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; gnomAD and population allele-frequency context; Ensembl VEP and transcript-aware variant consequences; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 11 October 2026: Official Human Phenotype Ontology portal; Official HPO repository and releases; The Human Phenotype Ontology in 2024; Encoding clinical data with HPO terms; 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 |
|---|---|---|---|
| Observe | What feature was actually present or assessed? | Clinical observation, examination and source note | A faithful finding |
| Encode | Which current HPO term best matches the observation? | Identifier, label, definition, synonyms and hierarchy | A versioned phenotype annotation |
| Compare | How does the whole profile relate to known profiles? | Present, absent, onset and frequency-aware similarity | A ranked hypothesis set |
| Validate | What evidence can confirm or weaken each hypothesis? | Genetic, biochemical, imaging and expert review | A bounded interpretation, not a diagnosis |
Inside this guide
1–12 · Foundations and design
- 1. Start with observations, not labels
- 2. Understand an ontology
- 3. Keep identifiers with labels
- 4. Move from broad to specific
- 5. Read definitions, not familiar words
- 6. Use synonyms for search, not identity
- 7. Separate present from assessed absent
- 8. Keep unknown separate
- 9. Add onset when known
- 10. Record severity carefully
- 11. Capture frequency at the right level
- 12. Retain the source and date
13–24 · Models and estimation
- 13. Handle measurements before categories
- 14. Avoid overcoding normal variation
- 15. Build a coherent profile
- 16. Understand semantic similarity
- 17. Specific terms can carry more information
- 18. Negative findings can refine a ranking
- 19. Expect incomplete disease annotations
- 20. Expect patient variability
- 21. Watch circularity
- 22. Do not equate rank with diagnosis
- 23. Combine genotype and phenotype transparently
- 24. Use current releases
25–36 · Diagnostics and interpretation
- 25. Audit obsolete terms
- 26. Protect privacy
- 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 observations, not labels
HPO represents phenotypic abnormalities such as findings from history, examination, imaging or laboratory work. A disease name is a different type of entity.
Working checkpoint — HPO term: Write the original observation before selecting a term. 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 an ontology
An ontology is more than a word list: it gives concepts identifiers, definitions and explicit relationships that software can traverse.
Working checkpoint — ontology hierarchy: Open the term page and inspect its definition and parents. 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. Keep identifiers with labels
Labels and synonyms can change while an HPO identifier provides a stable reference to the intended concept across systems.
Working checkpoint — phenotype annotation: Store the HP identifier and release alongside the human-readable 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. Move from broad to specific
A specific term inherits relationships to broader concepts. Recording only a broad parent loses information; choosing an unsupported child invents precision.
Working checkpoint — onset modifier: Select the most specific term justified by the source record. 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 definitions, not familiar words
Two terms may sound similar in everyday language but have different ontology definitions and diagnostic boundaries.
Working checkpoint — negated finding: Compare definitions and comments before choosing between near-neighbours. 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. Use synonyms for search, not identity
Synonyms help users find a concept, but the preferred identifier determines the encoded meaning.
Working checkpoint — semantic similarity: Map local wording to one reviewed concept and keep the original phrase. 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. Separate present from assessed absent
A feature that was checked and not found carries different information from one never assessed. Missing data must not become a negative phenotype.
Working checkpoint — HPO term: Use explicit negation only when the record supports it. 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. Keep unknown separate
Unknown, not recorded and not applicable are distinct states. Collapsing them can distort matching.
Working checkpoint — ontology hierarchy: Represent data state before any similarity calculation. 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. Add onset when known
A finding present at birth differs from the same feature appearing later. HPO annotations can be paired with onset information.
Working checkpoint — phenotype annotation: Record the best supported onset interval and its uncertainty. 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. Record severity carefully
Severity may be clinically useful, but local scales and subjective language do not automatically map to one universal value.
Working checkpoint — onset modifier: Preserve the original scale and avoid invented normalisation. 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. Capture frequency at the right level
Disease annotations can describe how often a feature appears among reported cases. Individual observations and disease-level frequencies answer different questions.
Working checkpoint — negated finding: Do not assign a disease frequency to one person. 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. Retain the source and date
A phenotype profile changes as a person is reassessed and as records are corrected. Provenance makes that change visible.
Working checkpoint — semantic similarity: Link every annotation to its note, observer and date. 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 measurements before categories
A numeric measurement contains more information than a broad abnormal/normal label, but interpretation depends on age, method and reference interval.
Working checkpoint — HPO term: Store the measurement, unit and reference before mapping a term. 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. Avoid overcoding normal variation
HPO focuses on abnormalities in disease contexts. Not every human difference is a pathological finding.
Working checkpoint — ontology hierarchy: Require a documented abnormal observation rather than coding assumptions. 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. Build a coherent profile
A useful profile is a set of independently supported findings, not a bag of attractive keywords.
Working checkpoint — phenotype annotation: Review whether terms duplicate, contradict or merely restate one observation. 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 semantic similarity
Similarity methods use ontology structure and annotation information to compare a profile with disease or gene profiles. Different algorithms can rank the same data differently.
Working checkpoint — onset modifier: Name the algorithm, reference annotations and release. 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. Specific terms can carry more information
A rare, specific term may contribute more to a similarity score than a common broad parent. This is a mathematical property, not proof of causation.
Working checkpoint — negated finding: Inspect which terms drive the ranking. 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. Negative findings can refine a ranking
Explicitly absent features may weaken some hypotheses, but age-dependent or incompletely assessed features need caution.
Working checkpoint — semantic similarity: Use only reliable negatives and preserve assessment context. 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. Expect incomplete disease annotations
Published disease profiles reflect reported cases and curation. Missing annotation does not prove a feature never occurs.
Working checkpoint — HPO term: Treat annotation absence as limited knowledge. 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. Expect patient variability
People with the same condition can have different features, severities and ages of onset. A non-perfect match can still be informative.
Working checkpoint — ontology hierarchy: Compare the whole pattern without demanding textbook completeness. 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. Watch circularity
If a disease annotation was built from the same case later used to test a method, apparent performance can be inflated.
Working checkpoint — phenotype annotation: Separate training, curation and evaluation cases. 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. Do not equate rank with diagnosis
A similarity rank organises hypotheses under a model. It is not a posterior probability unless the method explicitly and validly models one.
Working checkpoint — onset modifier: Report a ranked candidate list with assumptions. 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. Combine genotype and phenotype transparently
Phenotype matching can prioritise genes or variants, while genomic evidence addresses identity, frequency and consequence. Neither stream should silently overwrite the other.
Working checkpoint — negated finding: Keep phenotype and variant evidence in separate columns before integration. 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. Use current releases
Terms, definitions and annotations evolve. Old identifiers may be replaced or obsoleted with mappings.
Working checkpoint — semantic similarity: Pin the ontology and annotation release for reproducibility. 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 obsolete terms
An obsolete term should not be silently reused as if current. Replacement and reason-for-obsolescence notes preserve meaning.
Working checkpoint — HPO term: Resolve each obsolete identifier and log the mapping. 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. Protect privacy
Detailed phenotype combinations can be identifying, especially in rare disease. Sharing requires governance, minimisation and consent-aware practice.
Working checkpoint — ontology hierarchy: Publish only the detail appropriate to the approved purpose. 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
HPO supports structured description and computation; it does not replace clinical assessment or establish treatment.
Working checkpoint — phenotype annotation: Write “phenotypic similarity supports review of” rather than “proves”. 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 observable features into carefully defined groups shows why a shared vocabulary improves comparison.
Working checkpoint — onset modifier: Use plants, objects or fictional creatures rather than personal health data. 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
Genotype, environment, development and measurement all influence observed phenotype.
Working checkpoint — negated finding: Separate the observation from the proposed biological cause. 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
Hierarchies, sets, information content and similarity scores turn a vocabulary into a quantitative model.
Working checkpoint — semantic similarity: Change one term at a time and observe how rankings move. 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
Identifiers, graph edges, releases and APIs allow phenotype data to move between systems.
Working checkpoint — HPO term: Validate identifiers and reject silent label-only matches. 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
Phenotype informatics joins biology, medicine, ontology engineering and data stewardship and can illuminate STEM education and career pathways.
Working checkpoint — ontology hierarchy: Check current official programme and professional requirements separately. 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 very specific present finding can be informative, but an explicitly assessed absent finding can also help distinguish otherwise similar profiles.
Working checkpoint — phenotype annotation: Keep absent, unknown and unassessed 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 34 of 36
34. Write a reader-ready result
A strong summary names the observed terms, explicit negatives, onset, source, ontology release, comparison method and the limits of the ranked output.
Working checkpoint — onset modifier: Let the reader trace each encoded term back to evidence. 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 the observation, definition, specificity, negation and onset; pin the release; run sensitivity analyses; then seek independent clinical and molecular evidence.
Working checkpoint — negated finding: Strengthen annotation quality before strengthening interpretation. 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 phrases, mapped HPO IDs, present/absent state, modifiers, provenance, release files, algorithm settings, rankings and review notes.
Working checkpoint — semantic similarity: Make every phenotype-based result 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