eduKateSG · Why Science?
Give every cell type a stable identity—then keep measurement, state and biological context attached
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 shared identifier from an observation, a logical model from biological reality, and a useful comparison from a finished conclusion. The Cell Ontology, usually abbreviated CL, is a structured vocabulary of cell types. Its official OBO Foundry page says it was created in 2004 and is used by projects including HuBMAP, the Human Cell Atlas, CELLxGENE, the Single Cell Expression Atlas, BICCN, ENCODE and FANTOM5. CL links cell types to anatomy through Uberon, to biological processes through Gene Ontology relations, and to resources including ChEBI, Protein Ontology and PATO. Those links make annotations more comparable; they do not replace measured markers, morphology, developmental evidence or experimental context. This guide supports Primary Science, PSLE Science, Secondary Science, O-Level Science, O-Level Biology 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: Gene Ontology and traceable gene-function annotations; Protein Ontology and stable protein forms; NCBI Taxonomy and stable organism identifiers; OBI and reproducible investigations; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 13 October 2026: OBO Foundry Cell Ontology overview, checked 13 October 2026; Official Cell Ontology project documentation; Official Cell Ontology source repository; Main Cell Ontology OWL release; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science teaching and learning syllabus; 2026 MOE G2 Computing syllabus. Ontologies, definitions and release products can change, so preserve dated documentation, exact source records and file versions.
A deeper reading route
Cell-type names are deceptively difficult. Two laboratories may use different labels for similar cells, or the same label for cells measured with different markers and resolutions. CL offers stable identifiers so the intended class can travel with the data instead of depending on a local spreadsheet phrase.
The ontology is hierarchical. A specialised cell can be connected to broader parents, allowing a query to gather a narrow class or a wider family. The hierarchy is useful only when the selected class matches the evidence actually available.
CL can use logical definitions involving lineage, anatomy, molecular participants and capabilities. These axioms let software infer relationships, but the inference inherits the ontology version and every asserted premise.
Developmental lineage matters because cells that look similar at one moment may arise from different precursors. A develops-from relation records a biological proposition, not a direct observation unless the experiment measured lineage.
Anatomical context matters too. The official CL page explains that cell types are linked to Uberon through part-of relationships. A cell label without organ, tissue or compartment can be too vague for comparison.
Functional language requires care. A capable-of relationship links a cell type to a Gene Ontology biological process; it does not claim that every cell of that type performs the process continuously in every condition.
Single-cell RNA sequencing has made cell annotation highly visible, but expression profiles are measurements from samples, not identities by themselves. Reference mapping, clustering and marker scoring can support a CL term while batch, platform and sampling shape the result.
Cell state and cell type are not interchangeable. Activation, stress, cycling or disease may alter expression strongly without creating a new stable lineage-defined type. A responsible record keeps type, state and condition in separate fields.
A reproducible annotation stores the CL identifier, ontology release, mapping method, reference dataset, candidate alternatives, confidence and reviewer decision. Keeping only the displayed label removes most of the evidence needed for reinterpretation.
CL is community infrastructure rather than a finished dictionary. Terms, definitions and relationships can change as evidence and use cases evolve. Release pinning and migration notes therefore belong in a serious dataset.
Start at the evidence-supported level. If the experiment distinguishes only a broad immune-cell family, assigning a highly specific subtype creates false precision.
Keep original author labels beside CL mappings. Local language often preserves clues about the assay, tissue or gating strategy that a harmonised term cannot carry alone.
Treat marker panels as context-dependent evidence. One transcript or surface marker rarely defines a cell type universally across species, development and disease.
Do not turn a cluster number into biology automatically. Clustering depends on preprocessing, resolution and the sampled population; CL annotation is a separate interpretive step.
Use taxon and anatomy constraints as quality checks. An impossible species or tissue combination may reveal an identifier, sample or mapping error.
Record unresolved and mixed populations honestly. “Unknown”, “doublet suspect” or a broader parent can be scientifically stronger than a confident but unsupported subtype.
Validate interoperability with real queries. Test whether equivalent searches retrieve the intended cells across datasets, not merely whether every row contains an ontology ID.
Remember the happy purpose: shared identities let scientists compare cells across projects while preserving enough context to disagree constructively, test alternatives and improve the map together.
A classroom exercise can use labelled cards for cell type, organ, lineage and function. The lesson is the relationship among evidence fields, not medical classification.
| Stage | Question | Evidence | Responsible output |
|---|---|---|---|
| Identify | Which biological cell type is meant? | CL identifier, label and definition | A stable computable class |
| Contextualise | Where and how does the cell occur? | Anatomy, taxon, lineage, process and state | A bounded biological context |
| Annotate | What evidence supports the assignment? | Markers, morphology, function and method | A traceable cell-type annotation |
| Validate | Does the label survive review and new data? | Reference mapping, expert review and release comparison | A revisable classification, not automatic ground truth |
Inside this guide
1–12 · Foundations and description
- 1. Begin with the biological question
- 2. Use the stable CL identifier
- 3. Read the textual definition
- 4. Inspect parents and children
- 5. Trace developmental lineage
- 6. Add anatomical context
- 7. Add taxonomic context
- 8. Read capable-of relations
- 9. Connect molecular participants
- 10. Map local labels transparently
- 11. Handle synonyms carefully
- 12. Avoid label-only matching
13–24 · Relations, mappings and validation
- 13. Prepare single-cell evidence
- 14. Keep cluster and type separate
- 15. Distinguish type from state
- 16. Treat markers as panels
- 17. Use morphology when available
- 18. Use lineage data when available
- 19. Audit reference mapping
- 20. Check batch and platform effects
- 21. Detect doublets and mixtures
- 22. Respect rare-cell uncertainty
- 23. Pin the ontology release
- 24. Handle obsolete terms
25–36 · Learning, interpretation and pathways
- 25. Validate logical consistency
- 26. Test cross-dataset queries
- 27. Protect sensitive context
- 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 annotation note
- 35. Build a validation ladder
- 36. Finish with a reproducible bundle
Section 1 of 36
1. Begin with the biological question
Decide whether the task is atlas annotation, sample comparison, imaging, lineage analysis or data integration. The necessary resolution of cell identity changes with that question.
Working checkpoint — Cell Ontology identifier: Write the unit of analysis and the claim before opening a term browser. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 2 of 36
2. Use the stable CL identifier
A label can be renamed or reused, while a CL identifier supplies a durable reference to a class and its ontology history.
Working checkpoint — cell type: Store the full identifier as text, not only a spreadsheet label. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 3 of 36
3. Read the textual definition
The definition states the intended biological scope and often contains conditions that the short label does not reveal.
Working checkpoint — is-a hierarchy: Compare the definition with the evidence available in the sample. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 4 of 36
4. Inspect parents and children
The is-a hierarchy shows broader and narrower classes, helping an annotator choose the most specific justified level.
Working checkpoint — develops-from relation: Move upward when the evidence cannot distinguish sibling types. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 5 of 36
5. Trace developmental lineage
Some cell classes are partly characterised by the precursor from which they develop. Lineage assertions should not be inferred from similarity alone.
Working checkpoint — capable-of relation: Name whether lineage was measured, curated or assumed. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 6 of 36
6. Add anatomical context
Cells occur in tissues and compartments, and CL can connect to Uberon anatomy. Location can distinguish or constrain an annotation.
Working checkpoint — anatomical location: Record organ, tissue and compartment with stable identifiers where justified. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 7 of 36
7. Add taxonomic context
Species affects cell terminology, markers and anatomical interpretation even when a cross-species class exists.
Working checkpoint — Cell Ontology identifier: Keep NCBI Taxonomy identity beside every specimen. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 8 of 36
8. Read capable-of relations
A cell type may be linked to biological processes it is capable of performing. Capability is not continuous activity in every sampled cell.
Working checkpoint — cell type: Separate class capability from measured process activity. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 9 of 36
9. Connect molecular participants
Links to ChEBI and Protein Ontology can make molecular context explicit without forcing CL to own every chemical or protein concept.
Working checkpoint — is-a hierarchy: Resolve external identifiers and pin their releases too. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 10 of 36
10. Map local labels transparently
Legacy datasets often contain abbreviations, synonyms and laboratory-specific categories. A mapping table should retain the original and explain the crosswalk.
Working checkpoint — develops-from relation: Record exact, broad, narrow and uncertain mappings separately. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 11 of 36
11. Handle synonyms carefully
A synonym supports search but may be exact, broad, narrow or related rather than interchangeable in every context.
Working checkpoint — capable-of relation: Inspect synonym scope before replacing labels automatically. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 12 of 36
12. Avoid label-only matching
String similarity can suggest candidates yet miss homonyms, developmental qualifiers or species restrictions.
Working checkpoint — anatomical location: Require definition and context checks after automated matching. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 13 of 36
13. Prepare single-cell evidence
Single-cell annotation often combines clustering, marker expression, reference mapping and expert knowledge. Each component can disagree.
Working checkpoint — Cell Ontology identifier: Archive the matrix version, preprocessing and annotation method. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 14 of 36
14. Keep cluster and type separate
A computational cluster is a grouping under chosen parameters; it is not automatically one biological cell type.
Working checkpoint — cell type: Allow several clusters to share a type or one cluster to remain mixed. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 15 of 36
15. Distinguish type from state
Activation, cycling, hypoxia and stress can dominate expression while leaving lineage identity unchanged.
Working checkpoint — is-a hierarchy: Store type and state as separate annotation dimensions. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 16 of 36
16. Treat markers as panels
Markers vary with tissue, species, technology and condition. A multi-feature pattern is usually stronger than one favourite gene.
Working checkpoint — develops-from relation: Report positive, negative and ambiguous marker evidence. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 17 of 36
17. Use morphology when available
Imaging, shape and tissue position can support identity that expression alone cannot settle. The measurements must be linked to the same cell or population.
Working checkpoint — capable-of relation: Keep image provenance and scoring criteria with the annotation. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 18 of 36
18. Use lineage data when available
Tracing, fate mapping or developmental studies can test ancestry more directly than expression resemblance.
Working checkpoint — anatomical location: Do not relabel correlation as lineage observation. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 19 of 36
19. Audit reference mapping
A reference atlas can transfer labels only for cell types represented in that reference and at its resolution. Novel or shifted states may map poorly.
Working checkpoint — Cell Ontology identifier: Inspect distance, uncertainty and out-of-reference signals. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 20 of 36
20. Check batch and platform effects
Chemistry, depth, dissociation and laboratory handling can separate cells for technical reasons.
Working checkpoint — cell type: Test whether annotations track batch more strongly than biology. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 21 of 36
21. Detect doublets and mixtures
Two captured cells or contaminated profiles can express incompatible marker sets and attract a misleading hybrid label.
Working checkpoint — is-a hierarchy: Keep doublet scores and mixed evidence in the record. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 22 of 36
22. Respect rare-cell uncertainty
A tiny population may be real, undersampled or an artefact. Rarity increases the need for orthogonal evidence.
Working checkpoint — develops-from relation: Seek replication, imaging or targeted validation before naming a new type. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 23 of 36
23. Pin the ontology release
A later CL release may add, merge, obsolete or redefine terms. Exact reruns require the release or file hash.
Working checkpoint — capable-of relation: Store release date, PURL or checksum and retrieval date. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 24 of 36
24. Handle obsolete terms
Obsolete identifiers should remain traceable to replacement or consideration notes rather than being silently dropped.
Working checkpoint — anatomical location: Build a migration table and review every non-exact replacement. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 25 of 36
25. Validate logical consistency
Reasoners can detect unsatisfiable classes or infer parents from axioms, but correct logic still depends on correct inputs.
Working checkpoint — Cell Ontology identifier: Run consistency checks on the frozen release and imports. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 26 of 36
26. Test cross-dataset queries
Interoperability is demonstrated when queries recover intended comparable cells across studies with explainable exclusions.
Working checkpoint — cell type: Create a small gold-standard query set before scaling. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 27 of 36
27. Protect sensitive context
Cell data may derive from human participants, and ontology labels do not remove consent, privacy or re-identification risks.
Working checkpoint — is-a hierarchy: Apply governance and minimum-necessary disclosure to linked metadata. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 28 of 36
28. Connect to Primary Science
Classification, structure and function make a friendly bridge from observed similarities to organised biological groups.
Working checkpoint — develops-from relation: Use fictional cards and observable features rather than human samples. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 29 of 36
29. Connect to Secondary Science
Cells, tissues, differentiation and specialised function show how biological identity can be described at several levels.
Working checkpoint — capable-of relation: Ask which evidence supports each level in the hierarchy. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 30 of 36
30. Connect to mathematics
Sets, hierarchies, similarity scores and uncertainty show why boundaries can be useful without being perfectly sharp.
Working checkpoint — anatomical location: Use simple confusion matrices with invented cell labels. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 31 of 36
31. Connect to computing
Identifiers, graphs, imports and validation rules turn biological vocabulary into executable data infrastructure.
Working checkpoint — Cell Ontology identifier: Write assertions for missing IDs, duplicated cells and release drift. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Section 32 of 36
32. Connect to pathways
Cell annotation brings together biology, microscopy, genomics, statistics and software engineering.
Working checkpoint — cell type: Check current official institution pages before making pathway decisions. Inspect the definition and parents instead of trusting the visible label alone.
For scientific writing, prefer “annotated as” when the cell identity remains an interpretation.
Section 33 of 36
33. Did you know?
The OBO Foundry notes that CL has supported resources from Human Cell Atlas and HuBMAP to CELLxGENE, ENCODE and FANTOM5.
Working checkpoint — is-a hierarchy: Ask how one shared class enables comparison while local evidence stays visible. Compare marker, morphology, lineage and location evidence; one layer may be incomplete.
For pathways, connect biology, data standards and software without promising admission or employment.
Section 34 of 36
34. Write a reader-ready annotation note
A clear note names the CL term, original label, evidence, tissue, taxon, method, alternatives, release and reviewer.
Working checkpoint — develops-from relation: Use bounded language and put uncertainty in the main table. Separate measured expression from inferred identity and from temporary cellular state.
For students, ask which observations support the name and which parts came from a reference model.
Section 35 of 36
35. Build a validation ladder
Move from identifier and definition checks to marker panels, reference comparison, orthogonal measurement and independent replication.
Working checkpoint — capable-of relation: Stop the conclusion at the highest completed rung. Retain alternative candidates and confidence so later releases can be evaluated fairly.
For parents, the useful checkpoint is whether the classification can be traced back to actual evidence.
Section 36 of 36
36. Finish with a reproducible bundle
Deliver source labels, CL mappings, release files, sample metadata, analysis code, evidence panels, confidence and migration history.
Working checkpoint — anatomical location: Make every displayed cell type traceable to its sample and decision record. Keep the CL release, source label, tissue, taxon, assay and annotator beside the chosen class.
For computing, make invalid identifiers and taxon–anatomy conflicts fail loudly.
Contents · Previous section · Continue to the Science Learning Hub