How do you translate VCF variant files for a multilingual genetics team without changing alleles, genotype calls or reference positions? Variant Call Format, usually shortened to VCF, is a structured representation of sequence variants and associated evidence. It is not a spreadsheet whose column headings can be translated at will. A professional VCF localization project keeps machine-readable identifiers and biological claims stable while making interpretation, documentation, software interfaces and approved human-readable annotations understandable to readers in another language.
People searching for VCF translation, BCF localization, genomic variant annotation, VCF header INFO fields, FORMAT genotype fields, phasing, genome assembly identifiers and variant quality scores are often trying to solve several different problems. A physician may need a clear translated report. A bioinformatician may need a parser-safe localized annotation view. A laboratory may need to preserve a VCF supplied to a downstream pipeline. These deliverables have different permissions and acceptance tests; treating them as interchangeable can make a plausible translation scientifically wrong.
This world-facing guide explains VCF 4.5 concepts with practical worked examples, including CHROM, POS, REF, ALT, QUAL, FILTER, INFO, FORMAT, allele indices, phased genotypes, reference builds, symbolic structural variants and BCF. It concentrates on explaining existing evidence rather than interpreting an individual’s genetic health. A translated description must not be presented as a new variant call, clinical conclusion or substitute for qualified genetics and medical review. Use access controls appropriate to human genomic data and obtain the actual specification and pipeline documentation for each project.
The sixty-second answer
Keep the VCF or BCF’s reference positions, alleles, genotype indices, typed headers and record relationships intact. Translate explanatory prose, reports, tooltips and approved user-facing annotations in a separate layer. Check provenance, intended readers, data privacy, compatible versions and clinically sensitive claims. A fluent translation has failed if it changes what the underlying variant records assert.
1. VCF is a record of assertions about sequence variants
Each variant record connects a reference sequence location with reference and alternate allele descriptions, filtering outcomes, annotations and sometimes per-sample genotype data. The file is not an organism’s complete genome or a laboratory report. Its claims are conditional on the reference assembly and computational workflow.
Take a practical case: A team can write a French introduction explaining the purpose of its variant catalogue while retaining the chromosome positions and alleles exactly as delivered.
Watch for this failure: Mistaking a variant call for an established pathogenic finding collapses different levels of evidence. Rephrasing a computational prediction as a confirmed diagnosis is not translation.
A reliable check: Label the material as variant calls, identify the pipeline and reference build, and compare record counts and biological fields before and after localization.
2. The first line declares the format version
VCF uses a required fileformat meta-information declaration as its first header line, such as ##fileformat=VCFv4.5. The version is a compatibility clue for which record and header conventions a parser should expect, not a decorative document title.
Picture a real handover: A localized guide may explain in Arabic what VCF version 4.5 means; the literal version marker remains an interoperable machine-readable declaration.
An easy mistake is: Changing the marker to a localized sentence or claiming a newer format merely because the documentation was revised can break downstream reading.
To verify the result: Keep the original declaration byte-for-byte when the underlying file is not being migrated; verify compatibility with a standards-aware reader.
3. Metadata lines and declared fields
Meta-information lines beginning with two hash signs define fields, provenance and references. Structured INFO or FORMAT definitions have machine-readable IDs and typed attributes, while Description values often contain prose whose presentation may be localized outside the raw dataset.
For example, A Spanish glossary can render the Description for a depth field in Spanish and still preserve the ID DP, its Type and Number cardinality.
The risk is: Replacing DP with a translated abbreviation throughout a file is not a harmless glossary edit: applications often resolve the exact field ID.
Before sign-off: Inventory all declared IDs, differentiate display text from parser keys, and test that every body reference resolves to its original definition.
4. The header column names are structural
The mandatory tab-separated header identifies CHROM, POS, ID, REF, ALT, QUAL, FILTER and INFO. FORMAT and sample columns appear when genotype data are supplied. These tokens tell software where to find typed fields.
In an international workflow, A public-facing translated dashboard can label CHROM as chromosome in the user’s language while exporting the original #CHROM token in a VCF file.
The distinction matters because: Substituting localized labels into the raw header may make a file look welcoming to a human reader but invalid to an established VCF parser.
The acceptance test: Round-trip the file through a parser, confirm the mandatory columns and their order, and maintain translated display labels in a separate resource.
5. CHROM identifies a particular reference sequence
The CHROM field names the reference sequence containing the site. Values may be human chromosome names or accession-like identifiers, and they must match the chosen reference set rather than a translator’s preferred naming convention.
Take a practical case: A project has contigs named chr1 and chrM. The documentation can explain the mitochondrial contig, but the record identifiers must remain consistent with that reference.
Watch for this failure: Renaming chrM to an ordinary-language word or silently dropping a chr prefix may sever the connection between variants and the FASTA index.
A reliable check: Verify that CHROM values resolve to the declared contigs, compare sequence dictionaries, and record any deliberate build conversion as a separate data operation.
6. Reference assemblies cannot be guessed from chromosome names
The same chromosome label can occur in more than one assembly. Position numbers are meaningful only in the context of a particular reference sequence and build, including patch or alternate contig policy when relevant.
Picture a real handover: A curator translates a report discussing two human reference builds. She keeps the assembly accession or explicit build identifier beside each coordinate rather than implying they are interchangeable.
An easy mistake is: Conflating two assemblies can change the base at a reported position or point to a different region, while the translated sentence still sounds accurate.
To verify the result: Check header reference metadata, pipeline manifest and actual reference FASTA; do not infer assembly identity solely from an article’s country or language.
7. POS is a one-based reference position
VCF POS is conventionally a one-based position on the reference sequence. Interval notation used elsewhere may be zero-based and half-open, so copying coordinates into another format requires explicit conversion.
For example, A site at POS 100 occupies the hundredth reference base in VCF. An array offset of 99 may refer to the same base in a zero-based API.
The risk is: Translating an instruction that calls POS an offset from zero introduces an off-by-one error with consequences for annotation and cross-referencing.
Before sign-off: Test reference bases at several known loci and document coordinate-system conventions whenever data move from VCF into BED, browsers or code.
8. REF holds the reference allele
REF is the allele sequence observed in the chosen reference assembly at the stated locus. Its spelling is biological sequence information, not a phrase, abbreviation or translation unit.
In an international workflow, In the record REF=A and ALT=G, the letters name nucleotides; a German label may explain adenine and guanine but must not replace those tokens.
The distinction matters because: A helpful editor may expand A into a full translated word inside the field, making the record syntactically incompatible and changing its asserted meaning.
The acceptance test: Compare REF to the reference FASTA, preserve the exact sequence and length, and distinguish explanatory expansion from a modification of the allele.
9. ALT stores alternate alleles
ALT may contain one or more alternative allele descriptions, including nucleotide strings or permitted symbolic representations. In multiallelic records the order of ALT entries determines genotype allele indices.
Take a practical case: A genotype 1/2 refers to the first and second ALT alleles, not to a translator’s sorted list of alphabetic labels.
Watch for this failure: Resorting ALT values for display without remapping all dependent fields silently changes what a sample genotype, allele depth or frequency refers to.
A reliable check: Preserve ALT ordering; if any transformation is intentionally performed, validate dependent INFO and FORMAT arrays against their allele definitions.
10. Indels require anchored representations
Small insertions and deletions use VCF representations linked to a reference position, often including an anchoring base. Variant records can be biologically equivalent but syntactically different after normalization.
Picture a real handover: An insertion explained in prose as an added base may appear in a record whose REF and ALT both start with the same nucleotide.
An easy mistake is: Deleting the shared base during copyediting because it looks redundant can destroy the correct representation and the intended POS coordinate.
To verify the result: Keep the original REF/ALT pair, compare reference context and treat trimming or normalization as a documented bioinformatics operation.
11. Normalization and left alignment are not translation
Equivalent variants in repetitive sequence may be represented at different positions. Left alignment and parsimonious normalization are computational representation changes that must be validated against the reference.
For example, Two labs deliver different-looking indel records that normalize to the same biological change. A translator can explain the relationship without rewriting either source unilaterally.
The risk is: Claiming that normalization is a harmless text cleanup obscures reproducibility and can break source-to-target record traceability.
Before sign-off: Record the exact normalization tool, reference build and before-and-after variants when a scientific owner explicitly authorizes transformation.
12. ID fields are references, not prose
The ID column can hold known identifiers or a missing marker. A variant identifier may connect a record to a database entry, but its mere presence does not establish clinical significance.
In an international workflow, A reader sees an rs identifier next to a variant. Translate the surrounding explanation of database lookup, not the identifier’s characters.
The distinction matters because: Auto-correcting a database ID into a local spelling or expanding it as though it were a common English abbreviation can detach the reference.
The acceptance test: Confirm the identifier remains stable, test its target database where appropriate, and distinguish database membership from evidence strength.
13. QUAL is not a language-neutral percentage
QUAL conveys a Phred-scaled quality measure associated with a call under the producing pipeline’s model. It is not ordinarily a direct percentage of diagnostic certainty.
Take a practical case: A dashboard can explain in Japanese that larger QUAL values may represent stronger calling evidence, while qualifying the statement by algorithm and filter context.
Watch for this failure: Turning QUAL 30 into 30 percent confidence confuses a logarithmic score with a linear proportion and overstates what the metric means.
A reliable check: Preserve numeric values, define the producing caller, and inspect its documentation before supplying any probability interpretation.
14. FILTER distinguishes PASS, failures and missing state
The FILTER field communicates whether the record passed specified filters, failed named filters or has no recorded filtering status. PASS and the missing marker have different meanings.
Picture a real handover: A localized report may say that a record passed the declared filters while keeping the literal token PASS in the exported variant file.
An easy mistake is: Replacing missing FILTER information with PASS because both look blank to a reader manufactures a quality outcome.
To verify the result: Test three example records—PASS, a named failure and missing—and ensure the translated explanation preserves each state.
15. INFO contains structured site annotations
INFO is a semicolon-delimited collection of flags and keyed values attached to the variant record. The type and number of values are governed by header definitions, and some entries are not free-form words.
For example, A localized appendix can explain AC as allele count and AF as allele frequency without altering AC=2 or AF=0.125 in a real record.
The risk is: Converting every semicolon into punctuation for translated prose or replacing reserved keys with translated words breaks the data grammar.
Before sign-off: Use a parser to enumerate INFO keys and values, then cross-check their declared Type and Number with any human-facing descriptions.
16. Cardinality codes A, R and G have distinct meanings
VCF Number declarations can specify values per alternate allele, per allele including reference, or per genotype. Understanding the relationship is essential when a record contains multiple ALT entries.
In an international workflow, For two alternate alleles, a Number=A value may carry two entries; a Number=R value may carry three entries including REF.
The distinction matters because: A translation team that treats these counts as prose list lengths can accidentally shift allele-associated values by one place.
The acceptance test: Validate cardinality against the number of alleles and sample ploidy; do not reorder arrays to satisfy local punctuation preferences.
17. Allele frequency is contextual evidence
AF values refer to an allele frequency under a stated population, cohort or calling context. They are not always a universal property of the variant, and AC and AN may refer to the allele-count denominator.
Take a practical case: A population dataset says AF=0.02. Translate the cohort description and sample scope, not the claim into ‘two percent of all people worldwide’.
Watch for this failure: Confusing observed cohort frequency with a population-wide prevalence or a patient’s risk level changes the scientific assertion.
A reliable check: Retain cohort and filtering qualifiers; calculate displayed percentages only from the specific documented field meaning and source data.
18. INFO descriptions can be human-facing without moving keys
Some header Description strings contain useful explanatory English. They can be rendered in a multilingual schema glossary or UI tooltip, provided the canonical field identity remains intact.
Picture a real handover: A French tooltip explains that END denotes an end coordinate for a particular type of structural variant; the exported INFO key remains END.
An easy mistake is: Careless replacement of text inside a quoted structured header can introduce quote or delimiter syntax errors and cause parsing failures.
To verify the result: Prefer separate localization catalogues; if the source owner permits header prose edits, escape and validate the complete header.
19. FORMAT specifies per-sample field interpretation
FORMAT lists colon-separated field keys whose order governs the colon-separated values for each sample. GT, DP, GQ, AD and related fields can differ in presence by pipeline.
For example, In a sample row with GT:DP, the value 0/1:21 means a genotype index pair and a depth value; it is not a phrase with an optional number.
The risk is: Moving DP ahead of GT in the header while leaving sample values unchanged silently corrupts the interpretation of each column.
Before sign-off: Compare FORMAT keys with sample values for each row, including missing and trailing entries, under a version-appropriate parser.
20. GT uses allele indices, not translated allele names
The GT field refers to allele indices: zero indicates REF, positive integers select ALT entries, and missing calls are represented according to VCF syntax. The number of entries reflects ploidy.
In an international workflow, A diploid 0/1 sample is not automatically ‘normal plus disease’; it indicates one reference and one first-alternate allele in that call.
The distinction matters because: Expanding 1/1 into a disease diagnosis or reassigning index 1 after sorting ALT values transforms evidence into unsupported interpretation.
The acceptance test: Trace each genotype allele index to the exact REF/ALT entries and consult qualified clinical review for any health interpretation.
21. Phasing separators carry meaning
Slash-separated genotype calls are unphased in this convention; vertical bars indicate phased allele relationships. Phase sets and pipeline context help explain what is known about haplotypes.
Take a practical case: A localized table may say ‘phased’ for 0|1 and ‘unphased’ for 0/1 while preserving the original delimiter symbols.
Watch for this failure: Replacing the vertical bar with a slash because a local font renders it poorly destroys information about the stated phase.
A reliable check: Use literal genotype strings in exports; test phase-related fields and make explanatory prose explicit about uncertainty and scope.
22. Ploidy may vary across contexts
Genotype records may reflect different ploidy levels and missing alleles. One should not impose a diploid template on every chromosome, organism, sample or biological context.
Picture a real handover: A plant dataset includes multiple allele indices per call. The translated glossary should explain its ploidy rather than reducing each call to two alleles.
An easy mistake is: Assuming that every GT has exactly two values can truncate data or force imaginary missing values into valid records.
To verify the result: Check organism, sample metadata and caller conventions; validate genotype value counts where the scientific context defines expected ploidy.
23. Missing values mean absent information, not zero
VCF’s dot marker communicates missing or unspecified information in multiple fields. It does not by itself mean a negative test result, zero reads or the reference genotype.
For example, A translator explaining DP=. should distinguish ‘no recorded depth value’ from ‘depth was zero’ rather than choosing whichever phrase sounds natural.
The risk is: Replacing missing numeric measurements with 0 can change filtering, statistics and downstream clinical or research decisions.
Before sign-off: Audit missing markers by field and example; report data availability separately from a biological negative finding.
24. Genotype quality differs from site quality
GQ is commonly a per-sample genotype quality measure, whereas QUAL concerns the record or call in another sense. Both are influenced by the producing pipeline and must not be assumed equivalent.
In an international workflow, One sample has a low GQ at a high-QUAL variant. A good translation explains the distinction rather than promising the sample genotype is certain.
The distinction matters because: Treating GQ and QUAL as synonyms hides uncertainty specific to an individual sample and can mislead comparative results.
The acceptance test: Check the caller’s field definitions, preserve both measures and represent sample-level and site-level confidence separately.
25. Genotype likelihood arrays have defined ordering
PL and related fields describe genotype-likelihood information in a prescribed ordering, often with transformations of likelihood scale. Entries are indexed by genotype configuration, not a free-language list.
Take a practical case: A tri-allelic site can have several likelihood values whose ordering must match the allele and ploidy model.
Watch for this failure: Sorting likelihood entries by numeric size for attractive presentation without preserving their labels destroys their mapping to genotype hypotheses.
A reliable check: Use a standards-aware decoder and verify reconstructed genotype combinations before rendering human-facing explanations.
26. Sample identifiers are not freely localizable names
The columns following FORMAT identify samples and link to potentially sensitive provenance. Some names are pseudonyms; others may reveal a person, family or site.
Picture a real handover: A multinational consortium translates the consented explanatory material but retains protected sample keys in its controlled data environment.
An easy mistake is: Replacing sample IDs with translated personal names risks disclosure and can invalidate joins to manifests or external consent records.
To verify the result: Use approved pseudonymization and access controls; verify that each sample column still joins to its intended metadata and permissions.
27. BCF is a binary companion, not a translated VCF
BCF stores related variant information in a compact binary form and relies on declared header dictionaries and field types. Text rendering and binary serialization are separate operations.
For example, A research archive offers both .vcf.gz and .bcf exports with localized documentation. The biological content should match across representations.
The risk is: Editing BCF bytes with a text editor or assuming it contains translatable UTF-8 sentences is likely to corrupt the container.
Before sign-off: Use htslib-compatible tools to inspect or convert formats; test record and header equivalence after any authorized serialization.
28. Compressed VCF and tabix indexing
Block gzip compression and indexing enable efficient genomic-region access. An index is not a translation catalogue; it encodes offsets tied to a particular compressed file.
In an international workflow, An international data portal provides a translated description alongside a .vcf.gz file and its matching index for region queries.
The distinction matters because: Changing bytes in a compressed file without rebuilding its index may leave downloads possible but genomic-region searches unreliable.
The acceptance test: Recreate or verify indexes after any authorized rewrite and compare representative regional queries with a known-good reference.
29. Structural variants use special notation
VCF can represent larger events with symbolic alternate alleles, SVTYPE, END and other tags. These symbols are interpreted through definitions and cannot be expanded into ordinary text inside a data field.
Take a practical case: A translated report explains a deletion symbol and its associated interval while keeping the literal ALT and INFO syntax unchanged.
Watch for this failure: Replacing a symbolic allele with the full translated word for deletion or confusing END with a clinical study end date changes the record’s meaning.
A reliable check: Validate reference ranges, structural variant tags and symbolic allele definitions against the exact VCF version.
30. Breakend notation describes adjacency
Breakend representations can encode relationships between distant reference locations using bracket notation. Their grammar describes strand and junction relationships that simple prose cannot replace.
Picture a real handover: A curator explains a rearrangement in an accessible English caption while preserving the exact bracketed allele expression and both breakend partners.
An easy mistake is: Copyediting bracket order or treating chromosome labels as sentence punctuation can reverse or sever a structural relationship.
To verify the result: Round-trip breakend records through a supporting parser and compare both partner positions and orientations.
31. Copy-number descriptions need separate review
Copy-number variation descriptions may combine symbolic alleles, event length, depth estimates and sample-specific annotations. The word ‘gain’ can be more or less specific than a particular encoded call.
For example, A result summary says a segment has an estimated copy-number increase; the localized wording includes the assay and uncertainty qualifiers.
The risk is: Translating a threshold-dependent copy-number call into an unconditional genetic fact overstates the observation and potentially its clinical relevance.
Before sign-off: Keep assay method, copy-number scale, confidence and reference interval attached to any human-readable interpretation.
32. Variant-effect annotation depends on transcripts
Predicted consequences such as missense or splice-related effects are often computed against particular transcript models. Different transcript versions may give different labels for the same genomic change.
In an international workflow, A report has a variant annotated on two transcripts. The translation explains that the transcript-specific consequences differ instead of selecting one as definitive.
The distinction matters because: Omitting transcript accession versions or translating a predicted consequence as proven functional damage breaks the chain of evidence.
The acceptance test: Preserve transcript IDs and versions; compare annotation-tool outputs and use the exact source vocabulary for consequence labels.
33. Clinical interpretation is a separate responsibility
Database statements about pathogenicity, evidence strength or review status require specialist interpretation and current authoritative sources. A VCF alone rarely supports a patient-facing clinical recommendation.
Take a practical case: A genetic counsellor reviews a multilingual variant summary prepared from laboratory-approved findings, not from raw allele strings alone.
Watch for this failure: Conflating a variant’s presence with a diagnosis, or translating a category while dropping a negation or uncertainty label, creates real patient risk.
A reliable check: Require qualified clinical review for clinical materials and avoid releasing identifiable genomic records through general translation services.
34. Field-specific dictionaries beat global find-and-replace
A token can mean different things depending on whether it appears as CHROM, an INFO key, a quoted Description, sample ID or ordinary documentation. Global word substitution ignores the grammar.
Picture a real handover: The English word ‘pass’ appears in a report, but PASS inside FILTER is a reserved machine-readable state. The two contexts receive different treatment.
An easy mistake is: A search-and-replace pass over raw lines can accidentally alter all occurrences including keys, identifiers and quoted escaped characters.
To verify the result: Build a typed translation inventory with allowed editable locations, a glossary, locked tokens and a validation checklist.
35. Numeric formatting and units need caution
Localized reports may display decimal separators or units differently, but VCF data fields use machine-readable numeric syntax. Rendered values and stored values must be distinguished.
For example, A German-facing dashboard can show a frequency as 12,5% while keeping the original numeric representation in the exported VCF field.
The risk is: Writing a comma as a decimal separator inside an INFO value can be confused with the comma delimiter used for multi-valued attributes.
Before sign-off: Perform locale-specific number formatting only in the presentation layer; verify the stored values and exact cardinality.
36. Unicode, escaping and quoted descriptions
Human-facing header descriptions can contain punctuation and character encodings that software interprets under VCF rules. A translated sentence must not introduce unescaped quotes or forbidden separators.
In an international workflow, A translated Description field contains an apostrophe and a non-Latin term. It is tested as actual header syntax before being distributed.
The distinction matters because: A typographic quotation mark is not necessarily the same as a structural double quote; naive serialization can produce invalid metadata.
The acceptance test: Prefer separate multilingual documentation; if editing authorized descriptions, use library serializers and strict validation.
37. Round-trip testing is a scientific safeguard
A file may open in a text editor while losing biological fidelity. Round-trip checks should compare parsed field values, not only whether the export looks readable to a person.
Take a practical case: A team parses the source VCF, renders translations in an application and exports the same source call set. Alleles, positions and genotype arrays are compared.
Watch for this failure: A diff of rendered HTML proves nothing about whether an allele index, quality value or contig was altered by a background conversion.
A reliable check: Compare record counts, key columns, typed annotations and representative edge cases with machine-readable reports.
38. Privacy and governance accompany translation
Human genomics data can reveal sensitive health and family information. Translation workflows need appropriately authorized processors, retention controls, review access and provenance.
Picture a real handover: A multinational research group distributes localized training pages openly while granting the underlying identified variant files only to authorized collaborators.
An easy mistake is: Uploading raw VCF files into an unapproved language service for convenience may violate confidentiality obligations even if no fields are visibly translated.
To verify the result: Separate public educational examples from protected records, use synthetic data for demonstrations and audit all authorized transfers.
39. A safe end-to-end multilingual release
Define whether the deliverable is a report, UI, glossary, schema description or scientifically approved transformed dataset. Freeze a canonical data copy and version the translation artefacts.
For example, A release package contains the original VCF, its reference-build manifest, approved translated help text, a parser validation report and reviewer sign-off.
The risk is: Without a declared owner for biological fields, translators may feel pressured to ‘correct’ awkward data values that should instead be escalated.
Before sign-off: Assign responsibility for linguistic accuracy, data integrity and clinical claims separately; publish only after independent checks of all three.
40. Reading VCF through one coherent example
Consider an illustrative row for chromosome 1, POS 100, REF A, ALT G, FILTER PASS and a sample GT of 0/1. Each field answers a different question: location, allele hypothesis, filtering status and genotype indexing.
In an international workflow, An educator can translate this example into Hindi, Portuguese or Mandarin while keeping the variant tokens and reference assumption explicitly labelled as illustrative.
The distinction matters because: A casual summary saying ‘the patient has a disease allele at chromosome 1’ goes far beyond what the row alone establishes.
The acceptance test: Have a second reviewer reconstruct the exact source fields from the translated explanation and flag any new biological or clinical assertion.
41. Glossary design for international readers
Use a bilingual glossary with columns for protected field token, plain-language meaning, scope, example and ‘do not change’ status. Keep expansion of technical abbreviations separate from their code identity.
Take a practical case: A glossary entry for DP specifies whether it concerns site depth or sample depth in the producing schema, rather than promising one universal meaning.
Watch for this failure: A generic dictionary that maps every instance of a code to one translated phrase can be confidently wrong across callers and annotation sets.
A reliable check: Review glossary entries against real headers, source software documentation and at least one ambiguous field example.
42. When to consult other formats and references
VCF is downstream of sequencing, alignment and variant calling, but it is not a substitute for FASTQ quality records or SAM/BAM/CRAM read alignments. Different artifacts support different questions.
Picture a real handover: A reviewer comparing an unexpected variant can trace it to alignment evidence without pretending the final VCF itself stores every original sequencing read.
An easy mistake is: Translating a VCF description as if the file directly contained all raw reads misrepresents its provenance and limits.
To verify the result: Keep upstream data lineage visible and specify which file and workflow generated each reported value.
43. Version-sensitive claims belong in the release notes
The standards family evolves. A reader may encounter VCF 4.2, 4.3, 4.4 or 4.5, with differences relevant to particular tags or advanced representations.
For example, A reference page states which version it demonstrates and notes where a customer’s actual VCF header declares another version.
The risk is: Assuming that every example from a newer revision is valid in an older parser creates errors disguised as a translation problem.
Before sign-off: Record the version found in the source file and test against the actual receiving application’s supported specification.
44. Learning the difference between a label and a value
The core literacy skill is recognising when symbols are data versus language about data. GENOTYPE, REF and ALT may be explainable to a human reader, but their field values carry exact computational semantics.
In an international workflow, Students can annotate a synthetic record by colouring editable comments separately from immutable data tokens before producing a translated teaching note.
The distinction matters because: When a translation exercise encourages spelling changes inside a real data row, a learner can leave with a fluent but unsafe habit.
The acceptance test: Use synthetic examples for training; assess both clarity of explanation and ability to leave scientific records unchanged.
Six worked cases: applying the distinctions
Multiallelic calls
A fictional record uses REF=C and ALT=A,G with GT=1/2. The reader must understand that one sample genotype refers to the first and second alternates. Translating the explanation is permitted; swapping ALT order without recalculating GT and every allele-indexed field is not. Check that genotype 1/2 maps to A and G before and after the localization.
Missing versus filtered
A data manager receives three records: FILTER=PASS, FILTER=q10 and FILTER=.. A correct translation states that the first passed the declared checks, the second failed a named check, and the third has missing filter information. It does not conflate the third with a pass or translate q10 into an invented severity grade.
Conflicting reference builds
Two datasets report a variant at identical chromosome-position text while using different genome assemblies. A multilingual summary must include the reference identity and avoid presenting coordinates as directly comparable. Scientific reconciliation requires an authorized lift-over or variant normalization workflow, separate from ordinary document translation and reviewed for ambiguous mappings.
Clinical audience
A raw VCF includes a predicted effect label and a sample genotype. A family-facing report must not infer a diagnosis from those fields alone. Have the laboratory or qualified genetics team issue a validated interpretation, then translate the approved clinical document with rigorous management of negation, uncertainty, inheritance language and privacy.
Structural event
A symbolic deletion allele and END coordinate are explained in a multilingual figure caption. The encoded ALT syntax, sequence IDs and interval boundaries remain unchanged. Check that readers understand both the extent of the event and the limitations of the assay rather than assuming precise breakpoints from the visual label alone.
BCF export
An archive converts its canonical VCF to BCF for efficient retrieval. Localization happens in the portal’s interface and supporting notes. Compare parsed variants and typed annotations from both files, and do not mistake a binary conversion for a language translation. The link to the exact reference build belongs with both distribution files.
Questions readers ask
Can I translate CHROM, REF or ALT directly?
No. Those are structural fields or biological sequence values. Translate the interface explaining them, not the machine-readable keys or alleles.
Does a VCF genotype confirm disease?
No. A genotype call is not automatically a validated clinical interpretation. Appropriate laboratory evidence and qualified review are separate.
Is BCF a different language?
No. BCF is a binary data representation in the same technical ecosystem; it is not a natural-language edition of a VCF.
Why is 0/1 different from 1/1?
The first refers to one reference and one first-alternate allele in a diploid unphased call; the second refers to two first-alternate alleles.
What if header text really needs translation?
Prefer a presentation-layer glossary. If source owners permit changing Description strings, use a schema-aware serializer and validate the resulting file.
Can a decimal comma be used in a VCF numeric field?
Do not apply locale-specific number formatting to the stored field. Use localized formatting only when presenting numeric information to readers.
How do I prove a translation did not alter calls?
Compare parsed record identities, coordinates, alleles, filters, typed INFO values, genotype indices and sample links before and after.
Where should I read the standard?
Consult the GA4GH-maintained HTS specifications for the relevant VCF and BCF revisions, along with your actual variant-caller documentation.
Release checklist
Before releasing a multilingual variant resource, list the reference assembly and exact source file; capture a checksum and the declared VCF version; identify immutable column tokens and biological values; translate only authorized explanatory surfaces; compare typed fields through a parser; test multiallelic, missing, phased, structural and per-sample edge cases; protect personal genomic data; and obtain specialist review for any health-related conclusion. Record who approved the language and who approved the underlying scientific meaning.
Authoritative sources and further reading
GA4GH / hts-specs: VCF, BCF and related formats