eduKateSG · Why Science?
Turn genome coordinates into a connected map of genes, transcripts and features—while preserving the ontology, hierarchy and reference assembly
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. Generic Feature Format version 3, or GFF3, is a nine-column tab-delimited text format for describing features on biological sequences. The Sequence Ontology specification defines seqid, source, type, start, end, score, strand, phase and attributes; coordinates are positive one-based and inclusive, and feature types are constrained to Sequence Ontology terms or accessions. ID and Parent attributes can represent multi-level feature relationships. NCBI documents that its distributed GFF3 files follow the official specification while also using additional conventions, so consumers must read both the standard and producer documentation. 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: BED and genomic feature coordinates; tabix and indexed genomic text files; refget and globally identifiable reference sequences; Science Learning Hub; Education Hub; STEM Education; Careers by Subject Capability.
Primary and official sources checked on 12 October 2026: Official Sequence Ontology GFF3 specification; Sequence Ontology official resources index; NCBI documentation for its GFF3 annotation files and documented conventions; 2026 Singapore–Cambridge O-Level Biology syllabus; MOE G2/G3 Lower Secondary Science teaching and learning syllabus; 2026 MOE G2 Computing syllabus. Specifications, tools and recommendations can change, so preserve the dated documentation, exact source bytes, parser rules and software versions used.
A deeper reading route
GFF3 is a map of assertions, not merely a table of rectangles. Each row places a typed feature on a landmark sequence and may connect it to other features through identifiers. A gene can have transcript children; transcripts can have exon or CDS children; one child can sometimes belong to more than one parent. The relationships let software reconstruct biological models that no single interval can express.
The first eight columns have positional meanings, and the ninth is a structured attribute field. Tabs—not runs of spaces—separate columns. Undefined values use a dot. Reserved characters in attributes and some other fields require percent encoding under the specification. A parser that splits on whitespace or invents backslash escapes can corrupt names and relationships while leaving the row visually plausible.
Coordinates are one-based and inclusive. Start and end are positive integers, with start less than or equal to end under the circular-feature convention described by the specification. This differs from BED’s zero-based half-open model. A naïve copy between formats shifts boundaries or lengths; the conversion must also preserve strand, type, hierarchy and block meaning.
The type column is tied to the Sequence Ontology. Using a controlled term allows tools and people to distinguish gene, mRNA, exon, CDS and many other feature classes. Yet ontology conformance does not guarantee that an annotation is biologically correct. The source column, methods, evidence and release explain how the feature was produced.
ID identifies a feature within the file; Parent links a child to one or more parents. Multi-row features can repeat the same ID when one logical feature occupies disjoint regions. Ordering rows is therefore not a substitute for building an identifier graph. Duplicate incompatible IDs, missing parents or cycles can make a file parse but not model a coherent annotation.
CDS phase is often misunderstood. It applies to CDS features and indicates how many bases should be removed from the beginning of that feature, relative to its strand, to reach the next codon boundary. It is not the same as reading frame as a fixed genomic modulo, and it is not used to number exons. Validation should translate reconstructed CDS sequences under the declared genetic code.
Producer conventions matter. NCBI recommends accession.version sequence identifiers in its files and documents official and additional attributes, directives and model details. It also warns that its GFF3 output differs from the official specification in certain conventions. Interoperability therefore requires exact producer, release, assembly and consumer tests—not just a .gff3 extension.
Annotation comparison requires more care than matching gene names. Releases can split or merge models, move boundaries, change transcript sets, revise feature types or update stable accessions. A useful comparison records assembly compatibility, matches features by declared identifiers and genomic relationships, and reports additions, removals and modifications separately. Counts alone can hide a wholesale remodelling in which the total happens to remain similar.
Subsetting can also damage meaning. Selecting rows from one interval may omit parents outside the window, shared children, directives, sequence-region declarations or the embedded sequence tail. A subsetter should declare whether it closes the Parent graph, rewrites IDs or keeps dangling references. Validation after subsetting is not optional because a syntactically neat regional file may no longer represent a coherent biological model.
For students, GFF3 is a cheerful example of structured explanation. Coordinates say where something is; a controlled term says what kind of feature it is; attributes and Parent links say how parts belong together. The broader lesson is that good science needs both measurements and relationships, with every assumption visible enough to test.
| Stage | Question | Evidence | Responsible output |
|---|---|---|---|
| Locate | On which landmark and coordinates is the feature? | seqid, start, end and strand | A reference-relative interval |
| Classify | What kind of feature is asserted? | source, Sequence Ontology type and score | A typed annotation claim |
| Connect | How do parts form larger models? | ID, Parent and multi-row relationships | A feature hierarchy |
| Validate | Does the annotation match reference and method? | Assembly, schema, ontology, sequence checks and provenance | A trustworthy annotation set |
Inside this guide
1–12 · Foundations and design
- 1. Start with the annotation question
- 2. Use the official specification
- 3. Recognise nine tab-delimited columns
- 4. Split on tabs, not spaces
- 5. Use dots for undefined fields
- 6. Read seqid as a landmark
- 7. Prefer stable sequence identifiers
- 8. Read source as method context
- 9. Use Sequence Ontology types
- 10. Read start and end as one-based inclusive
- 11. Handle circular landmarks
- 12. Interpret score by source
13–24 · Models and estimation
- 13. Interpret strand carefully
- 14. Interpret phase only for CDS
- 15. Parse attributes structurally
- 16. Percent-encode reserved characters
- 17. Use ID within its scope
- 18. Use Parent to build hierarchy
- 19. Allow shared children when valid
- 20. Handle multi-row features
- 21. Check cycles and incompatible IDs
- 22. Respect directives
- 23. Handle comments separately
- 24. Understand the FASTA tail
25–36 · Diagnostics and interpretation
- 25. Compare carefully with BED
- 26. Index only sorted compatible files
- 27. Read NCBI conventions explicitly
- 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. Start with the annotation question
A GFF3 file may describe genes, transcripts, regulatory elements, alignments or other sequence features. Begin with producer, assembly, release and intended use.
Working checkpoint — GFF3: Write the annotation scope before interpreting rows. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 2 of 36
2. Use the official specification
The Sequence Ontology publishes the GFF3 specification and resource links. Producer-specific files can add documented conventions without becoming universal rules.
Working checkpoint — genome annotation: Pin the specification snapshot and producer documentation. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 3 of 36
3. Recognise nine tab-delimited columns
Each ordinary feature line has seqid, source, type, start, end, score, strand, phase and attributes separated by tabs.
Working checkpoint — seqid: Reject rows with the wrong column count. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 4 of 36
4. Split on tabs, not spaces
The official specification permits unescaped spaces within fields and requires parsers to use tabs as delimiters.
Working checkpoint — Sequence Ontology: Preserve internal spaces while validating tabs. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 5 of 36
5. Use dots for undefined fields
A dot represents an undefined value in applicable columns. Missing does not mean zero, false or biologically absent.
Working checkpoint — ID and Parent: Carry unknown states explicitly into derived tables. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 6 of 36
6. Read seqid as a landmark
seqid identifies the sequence or landmark used for coordinates. It must match the intended assembly and reference naming system.
Working checkpoint — feature hierarchy: Resolve every seqid against a versioned sequence manifest. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 7 of 36
7. Prefer stable sequence identifiers
NCBI uses accession.version in its GFF3 output and recommends it over ambiguous labels such as chr1 for unambiguous sequence identity.
Working checkpoint — GFF3: Store accession, assembly and sequence digest together. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 8 of 36
8. Read source as method context
The source column names an algorithm, procedure or database associated with feature generation. It helps qualify the ontology type.
Working checkpoint — genome annotation: Keep method documentation and software release beside the file. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 9 of 36
9. Use Sequence Ontology types
The type column must be a Sequence Ontology term or SO accession representing sequence_feature or a descendant under the official specification.
Working checkpoint — seqid: Validate terms against the relevant ontology release. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 10 of 36
10. Read start and end as one-based inclusive
GFF3 coordinates are positive one-based integers and both endpoints are included. A one-base feature has start equal to end.
Working checkpoint — Sequence Ontology: Test first-base and one-base examples. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 11 of 36
11. Handle circular landmarks
For features crossing the origin of a circular sequence, the specification permits an end beyond the landmark length to preserve start less than or equal to end.
Working checkpoint — ID and Parent: Use reference length and circularity metadata during normalization. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 12 of 36
12. Interpret score by source
The score column is a floating-point value when defined, but its biological meaning depends on the feature type and producing method.
Working checkpoint — feature hierarchy: Do not compare scores without compatible definitions. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 13 of 36
13. Interpret strand carefully
The strand column can be +, -, . or ? under the format rules. Unknown and not-applicable states should not be collapsed.
Working checkpoint — GFF3: Preserve strand through coordinate conversions. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 14 of 36
14. Interpret phase only for CDS
Phase is 0, 1 or 2 for CDS features and a dot otherwise. It indicates how translation continues from the beginning of that CDS segment on its strand.
Working checkpoint — genome annotation: Reconstruct CDS in strand order before validating translation. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 15 of 36
15. Parse attributes structurally
Column nine contains semicolon-delimited tag-value pairs with equals signs and comma-separated multiple values under defined escaping rules.
Working checkpoint — seqid: Use a GFF3 parser rather than ad hoc string splitting. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 16 of 36
16. Percent-encode reserved characters
Tabs, newlines, carriage returns, percent and control characters must be encoded, and attribute-reserved punctuation needs encoding in values.
Working checkpoint — Sequence Ontology: Decode exactly once and reject invalid escapes. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 17 of 36
17. Use ID within its scope
ID identifies a feature within the file and is required when a feature has children or spans multiple rows. It is not automatically a stable external accession.
Working checkpoint — ID and Parent: Distinguish file-local IDs from database identifiers. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 18 of 36
18. Use Parent to build hierarchy
Parent points from a child feature to one or more parent IDs. It supports gene-transcript-exon relationships beyond a two-level table.
Working checkpoint — feature hierarchy: Resolve all parents and report missing targets. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 19 of 36
19. Allow shared children when valid
The specification allows a feature such as an exon to belong to more than one parent. Tree-only code can duplicate or lose shared structure.
Working checkpoint — GFF3: Represent relationships as a graph where necessary. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 20 of 36
20. Handle multi-row features
Several rows with the same ID can represent one feature occupying disjoint regions. Their attributes should remain compatible.
Working checkpoint — genome annotation: Group by ID only after validating type, seqid and relationships. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 21 of 36
21. Check cycles and incompatible IDs
Parent relationships should form meaningful biological structures. Cycles, conflicting types or repeated IDs with incompatible locations signal errors.
Working checkpoint — seqid: Run graph-level validation after row parsing. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 22 of 36
22. Respect directives
Double-hash directives such as ##gff-version and ##sequence-region convey file-level context; producer-specific directives need documented handling.
Working checkpoint — Sequence Ontology: Preserve directives with every subset. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 23 of 36
23. Handle comments separately
Single-hash comment lines are not feature rows but can carry useful human context. They should not be parsed as nine-column annotations.
Working checkpoint — ID and Parent: Retain comments when provenance depends on them. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 24 of 36
24. Understand the FASTA tail
GFF3 can include a ##FASTA directive followed by embedded sequence records. Feature parsing ends and FASTA parsing begins at that boundary.
Working checkpoint — feature hierarchy: Verify embedded sequences against seqids and coordinates. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 25 of 36
25. Compare carefully with BED
BED uses zero-based half-open intervals and a different field model. Converting GFF3 requires coordinate change plus decisions about hierarchy, type and attributes.
Working checkpoint — GFF3: Validate lengths and relationships after conversion. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 26 of 36
26. Index only sorted compatible files
Position-sorted bgzip-compressed GFF3 can be indexed with tabix under appropriate parser rules. The index does not validate feature hierarchy.
Working checkpoint — genome annotation: Test indexed results against a sequential parser. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 27 of 36
27. Read NCBI conventions explicitly
NCBI documents its official and unofficial directives, accession.version seqids and additional attributes and notes differences from the official specification.
Working checkpoint — seqid: Use producer documentation for producer-specific semantics. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 28 of 36
28. Connect to Primary Science
A habitat map can label observations by place, type and parent group, showing that location and classification answer different questions.
Working checkpoint — Sequence Ontology: Use fictional features and simple hierarchies. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 29 of 36
29. Connect to Secondary Science
Genes and transcripts become easier to reason about when students separate genomic location, feature type, strand and relationships.
Working checkpoint — ID and Parent: Avoid presenting one annotation model as the only possible biology. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 30 of 36
30. Connect to mathematics
Inclusive intervals, graphs, ordered segments and coordinate conversions connect GFF3 to discrete mathematics.
Working checkpoint — feature hierarchy: Calculate feature length and trace one parent graph. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 31 of 36
31. Connect to computing
Delimited parsing, percent encoding, ontologies, graph validation and indexed access make GFF3 a practical data-engineering lesson.
Working checkpoint — GFF3: Test malformed escapes, missing parents and circularity. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Section 32 of 36
32. Connect to pathways
Genome annotation connects biology, ontology engineering, software, curation and data stewardship and can illuminate STEM pathways.
Working checkpoint — genome annotation: Check current official qualifications separately. Read every interval beside its feature type, strand, Parent graph, method and reference sequence.
For future study, connect genetics to graphs, ontologies and software while checking pathways separately.
Section 33 of 36
33. Did you know?
GFF3 can represent more than one level of feature hierarchy, so a gene can connect to transcripts and then exons or CDS rows through ID and Parent links.
Working checkpoint — seqid: Draw the graph instead of reading rows in isolation. Preserve directives, comments, unknown values and rejected rows; a flattened table is not the full annotation.
For scientific writing, report producer conventions, hierarchy checks, transformations and limits.
Section 34 of 36
34. Write a reader-ready annotation note
Name producer, assembly, release, specification, ontology, directives, conventions, validation tools, transformations and source hashes.
Working checkpoint — Sequence Ontology: Keep a seqid-to-reference manifest beside the file. Separate syntactic validity, ontology conformance, model coherence and biological evidence.
For students, ask how location, type and family relationships answer different questions.
Section 35 of 36
35. Build a validation ladder
Check tabs, columns, encoding, coordinates, ontology types, IDs, Parent graph, CDS phase, reference agreement and indexed equivalence.
Working checkpoint — ID and Parent: Stop before analysis when hierarchy or reference remains ambiguous. Treat seqids, coordinate convention, percent encoding and feature relationships as scientific inputs.
For parents, ask whether the map is tied to a named assembly and annotation release.
Section 36 of 36
36. Finish with a reproducible bundle
Deliver original and transformed hashes, schema notes, reference manifest, ontology snapshot, commands, validation logs and bounded biological claims.
Working checkpoint — feature hierarchy: Make every feature traceable to exact rows, sequence and method. Keep source hash, producer, assembly, release, specification snapshot, ontology version and validation report together.
For a classroom model, build an invented feature graph without personal data.
Contents · Previous section · Continue to the Science Learning Hub