Skip to main content Link Menu Expand (external link) Document Search Copy Copied
Biology Skills

Qualifying variants

Summary

A qualifying variant is a variant that satisfies a declared qualification rule under a defined genomic analysis context.

Qualification is not an intrinsic property of a variant.

The same variant can qualify under one analysis specification and not qualify under another because the purpose, inputs, thresholds, reference resources, or required evidence differ.

A qualifying variant set (QV set) makes these criteria explicit as a structured, versioned specification rather than leaving them embedded inside pipeline code.

This is important because variant-selection criteria are part of the scientific method. They determine which genomic records enter subsequent QC, association testing, interpretation, review, or reporting workflows.

The Qualifying Variant Set Standard (QVSS) defines a minimal interoperable representation for these criteria.

Core rules

  • Treat variant qualification as contextual, not intrinsic.
  • A variant qualifies only relative to a declared QV set, version, application context, input data, and evaluation.
  • Make thresholds, logical conditions, resources, and missing-value behaviour explicit when they determine variant qualification.
  • Do not hide scientifically consequential selection criteria only inside scripts, command-line arguments, notebooks, or undocumented pipeline defaults.
  • Distinguish a QV set from the software that evaluates it. The specification defines criteria; an implementation executes them.
  • Distinguish individual rule outcomes from final set-level qualification.
  • Do not call a variant a qualifying variant unless a set-level qualification rule is defined and satisfied.
  • Do not equate qualification with retention, exclusion, prioritisation, pathogenicity, reportability, statistical significance, causality, or clinical action.
  • Preserve the semantics and provenance of fields used by qualification rules.
  • Preserve reference genome, coordinate, transcript, interval, annotation, population, or external-resource identity whenever these affect rule evaluation.
  • Handle missing values explicitly. Do not silently convert missing or unresolved values into pass or fail.
  • Preserve unknown outcomes rather than forcing binary results when the declared evidence is insufficient.
  • Do not silently coerce incompatible datatypes or reinterpret field values.
  • Use explicit profiles or extensions when a rule requires semantics beyond simple deterministic comparisons.
  • Preserve stable QV-set identifiers, versions, rule identifiers, and relevant checksums.
  • Treat any change to normative qualification logic as a new QV-set version.
  • Record which QV set was actually applied to an analysis.

Qualification is contextual

Consider a variant with population allele frequency:

AF = 0.02

An illustrative common-variant study might define:

AF >= 0.01

and the variant could qualify.

A rare-disease candidate set might instead require:

AF < 0.001

and the same variant would not qualify.

Nothing about the underlying variant changed.

The qualification context changed.

Therefore avoid statements such as:

This is a qualifying variant.

without enough context to answer:

Qualifying under which QV set?
Which version?
For what analysis?
Using which inputs and resources?

QV set, rule outcome, and qualifying variant

Keep three concepts separate.

QV set

A structured, versioned specification containing the rules used to evaluate variants.

Conceptually:

QV set
├── identity and version
├── application context
├── declared inputs
├── rules
├── resources
├── missing-value behaviour
├── profiles / extensions where required
└── optional qualification rule

Rule outcome

The result of evaluating an individual rule.

For example:

minimum_depth = satisfied
population_frequency = not_satisfied
panel_overlap = unknown

A rule outcome does not automatically determine whether the variant qualifies.

Qualifying variant

A variant satisfies the QV set’s declared final qualification rule under the actual application context.

If the QV set does not define a final qualification rule, it may produce rule outcomes without defining a single binary qualifying status.

Do not invent a final qualifying state from individual rules when the specification does not define one.

Criteria are scientific parameters

Selection criteria can alter the population of variants entering every downstream analysis.

Examples include:

  • variant or genotype quality
  • read depth
  • call rate
  • allele frequency
  • Hardy-Weinberg thresholds
  • genomic-region overlap
  • gene-panel membership
  • consequence annotations
  • database flags
  • inheritance-related predicates
  • structural-variant properties
  • previously computed annotations or model outputs

These values should not be treated as incidental software settings when changing them can alter the scientific result.

A workflow such as:

bcftools view ...

or:

plink --maf 0.01 ...

does not by itself provide a durable scientific specification unless the relevant qualification logic and context remain recoverable.

Field semantics

A field name is not necessarily a complete definition.

Ambiguous depth

This rule is under-specified:

DP >= 10

because DP could refer to:

site-level INFO/DP
sample-level FORMAT/DP
derived depth
a particular member of a family
another implementation-specific field

Prefer an explicitly defined input whose source and scope are known.

Ambiguous allele frequency

This rule is under-specified:

AF < 0.01

unless the analysis establishes what AF represents.

Relevant distinctions may include:

cohort allele frequency
global reference-population frequency
population-specific frequency
maximum population frequency
allele count-derived estimate
another defined source

The threshold and the quantity being thresholded are both part of the scientific rule.

Ambiguous interval overlap

This rule is incomplete:

variant overlaps disease_panel.bed

if the following are unknown:

reference assembly
coordinate convention
interval semantics
minimum required overlap
panel release
resource checksum or immutable identity

Use the relevant genomic-reference and interval rules before treating the overlap as reproducible.

Missing values and unknown outcomes

Missing information is not automatically failure and is not automatically success.

A QV rule should declare what happens when a required field is missing.

Possible behaviours can include:

fail
pass
unknown
error

When no missing-value behaviour is declared under QVSS core semantics, treat the result as unknown.

For example:

rule:
  population_af < 0.001

observed input:
  population_af = missing

must not silently become:

population_af = 0

and must not automatically imply:

rare variant = true

unless the declared rule explicitly defines that behaviour.

Unknown values must also remain unknown through logical operations when the result cannot otherwise be determined.

For example:

not(unknown) = unknown

Do not use Boolean coercion that converts unknown evidence into false certainty.

External resources

Many qualification rules depend on external resources.

Examples include:

gene panels
BED interval sets
population-frequency databases
clinical databases
annotation releases
transcript sets
blacklists
reference genomes

The rule:

overlaps current disease panel

is not reproducible if “current disease panel” changes over time.

Preserve sufficient resource provenance to identify the resource used during the actual analysis.

Where appropriate, preserve:

resource identifier
version / release
reference assembly
retrieval date
checksum

Changing a normative external resource can change qualification even when the visible rule text is unchanged.

Complex genomic rules

Not every genomic criterion is a simple per-variant comparison.

Examples include:

compound heterozygosity
phase-dependent inheritance
multi-variant genotype configurations
family segregation
grouped variant counts
multi-record structural events

Do not reduce these to a misleading atomic expression merely to fit a simple rule format.

If the complex result has already been computed upstream, it can be supplied as an explicitly defined input predicate.

If the computation itself is part of the qualification semantics, use a declared profile or extension that defines the required behaviour.

The implementation must not invent hidden semantics.

Multi-stage qualification

QV criteria can be applied at several stages of one genomic workflow.

For example:

called variants
      ↓
QV set A: technical quality
      ↓
analysis-ready variants
      ↓
QV set B: study eligibility
      ↓
candidate variants
      ↓
QV set C: disease-specific criteria
      ↓
variants for downstream interpretation

A variant can therefore:

qualify under QV set A

and:

not qualify under QV set C

without inconsistency.

Each statement answers a different genomic question.

Do not collapse sequential QV sets into one unexplained label such as:

filtered variant

when the distinction affects reproducibility or interpretation.

Versioning and audit

A QV set should have stable identity independent of the filename used to store it.

Preserve as applicable:

QVSS version
QV set identifier
QV set version
QV set title
rule identifiers
checksum
declared profiles or extensions
external resource provenance

A published QV set should be treated as immutable.

Changing any normative component, including:

threshold
operator
rule statement
logical structure
missing-value behaviour
profile
normative extension

creates a different QV-set version.

Do not edit a released QV set in place while retaining the same version.

QV application record

The specification and its use are separate objects.

A QV application record identifies which specification was actually applied.

Conceptually:

qv_application:
  qv_set_id: rare_disease_candidates
  qv_set_version: 2.1.0
  qvss_version: 1.0
  qv_set_checksum: <digest>
  applied_at: <timestamp>
  implementation:
    name: <software>
    version: <version>
  input_data: <dataset identifier>

This record supports questions such as:

Which criteria were applied to this genome?
Which version was used?
Can I retrieve the exact rule set?
Was the same QV set used for another analysis?

It does not assert that a resulting variant is pathogenic, causal, reportable, or clinically actionable.

Relationship to qualifying evidence

Qualifying variants and qualifying evidence solve different problems.

Qualifying variants

Ask:

Does this genomic variant satisfy the declared criteria for this analysis context?

Conceptually:

variant
  +
QV set
  ↓
rule outcomes
  ↓
optional qualifying status

Qualifying evidence

Asks:

Is verifiable evidence available under a declared evidence rule?

Conceptually:

evaluated item
  +
evidence rule
  ↓
raw evidence outcome
  ↓
evidence-availability representation

Do not assume the Boolean semantics are interchangeable.

In QVSS, a satisfied rule can be represented as a positive rule outcome.

In QEM, raw TRUE is deliberately defined as detection of a contradiction, absence, or weakening signal and therefore maps differently.

Never copy QVSS rule outcomes directly into a QEM matrix without an explicitly defined evidence rule and mapping.

AI behaviour

Before describing a variant as qualifying:

  1. Identify the genomic analysis context.
  2. Identify the QV set.
  3. Verify the QV-set version.
  4. Determine whether the QV set actually defines a final qualification rule.
  5. Identify the variant representation and relevant genomic reference.
  6. Resolve the declared input fields and their meanings.
  7. Check the operators, values, and logical structure.
  8. Check the declared missing-value behaviour.
  9. Identify external resources and their relevant versions.
  10. Determine whether profiles or extensions are required.
  11. Evaluate unknown values without silently coercing them.
  12. Preserve the resulting rule outcomes separately from downstream interpretation.
  13. Record the QV application sufficiently to identify which criteria were actually used.

Do not infer that a variant is:

pathogenic
causal
significant
reportable
actionable

merely because it qualifies.

When reproducing an analysis, use the exact declared QV-set version and external resources rather than recreating the criteria from prose or memory.

Common failure modes

Qualification treated as an intrinsic variant property

A database field says:

qualifying_variant = true

but does not identify the criteria.

The statement is incomplete.

Qualification requires a declared context and QV set.

Hard-coded scientific criteria

A script contains:

if AF < 0.001 and DP >= 10:

but the thresholds are not represented anywhere outside the code.

The software may be reproducible while the scientific selection protocol remains difficult to discover, compare, audit, or reuse.

Missing value treated as passing

A rare-variant rule requires:

AF < 0.001

but AF is unavailable.

The system substitutes zero and the variant qualifies.

This converts missing information into evidence of rarity.

Ambiguous field source

Two implementations evaluate:

DP >= 10

One uses INFO/DP.

The other uses sample FORMAT/DP.

Both claim to implement the same criterion.

They are not necessarily evaluating the same quantity.

Unversioned gene panel

A QV set says:

overlaps paediatric disease panel

but does not identify which release of the panel was used.

The panel changes later.

The analysis can no longer be reconstructed reliably.

Qualification treated as pathogenicity

A variant satisfies all candidate-selection criteria.

The system reports:

pathogenic variant

Qualification only establishes that the declared criteria were satisfied.

Pathogenicity requires a separate interpretive process.

Qualification treated as an action

A variant satisfies the QV set and is automatically reported clinically.

QVSS qualification does not itself specify whether a variant should be retained, excluded, ranked, interpreted, reported, or acted upon.

Complex inheritance reduced to a per-variant filter

A compound-heterozygous model requires two variants, phase, and appropriate inheritance context.

The system represents this as:

compound_het = variant_A == true

without defining how the pair or phase was established.

Complex multi-record semantics require explicit computation and provenance.

Released criteria edited in place

A QV set version 1.0.0 changes:

AF < 0.01

to:

AF < 0.001

without changing its version.

Historical analyses referring to 1.0.0 are now ambiguous.

Examples

Example 1: the same variant in different analyses

Suppose:

variant population AF = 0.02

An illustrative GWAS QV set requires:

AF >= 0.01

Result:

qualifies for this GWAS criterion

An illustrative rare-disease QV set requires:

AF < 0.001

Result:

does not qualify for this rare-disease criterion

The genomic variant has not changed.

The analysis question has.

Example 2: sequential QV sets

A variant has:

QUAL = 80
sample DP = 35
population AF = 0.005

An upstream technical QV set might require:

QUAL >= 30
DP >= 10

and the variant qualifies.

A downstream rare-disease QV set might require:

population AF < 0.001

and the variant does not qualify.

Do not summarise both results simply as:

variant passed filtering

Record which criteria were satisfied at which stage.

Example 3: missing population frequency

Rule:

population AF < 0.001
missing: unknown

Variant:

population AF = missing

Result:

rule outcome = unknown

not:

rule outcome = true

Missing population data are not evidence that the variant is rare.

Example 4: panel overlap

Rule:

variant overlaps disease panel

A reproducible implementation must establish enough context to interpret that statement, including the relevant genomic reference and the identity of the panel resource.

Two analyses using different panel releases can legitimately produce different qualifying sets.

Authoritative standards

The Qualifying Variant Set Standard (QVSS) defines the normative representation of qualifying variant criteria.

Biology Skills explains the biological and computational distinctions that an AI must preserve when designing or applying such criteria. It does not replace the current QVSS specification.

Use the current QVSS standard for exact:

  • required fields
  • rule structure
  • logical operators
  • comparison operators
  • datatype behaviour
  • missing-value semantics
  • three-valued logic
  • profile and extension requirements
  • conformance requirements
  • registry and application-record requirements

QVSS defines how qualification criteria are represented.

It does not establish whether a particular threshold, resource, gene panel, biological assumption, or clinical criterion is scientifically correct.

Sources

  • Lawless D, Saadat A, Ait Oumelloul M, et al. Application of qualifying variants for genomic analysis. Bioinformatics. 2026;42(2):btaf676. https://doi.org/10.1093/bioinformatics/btaf676
  • Swiss Genomics Association. Qualifying Variant Set Standard. SGA-QVSS-1.0. 2026.