Glossary

A result is written in the registries’ own words, because paraphrasing what a registry said is how a tool ends up reporting something nobody said. The cost is a page of LOEUF, missense_variant and criteria provided, multiple submitters, no conflicts. Here is what all of it means.

On a result, every one of these terms is underlined: hover it, tap it, or tab to it for the same definition in place. The switch above the result lists them under each section instead, and every export carries the definitions for the terms it uses.

These definitions explain what each term means in general, each with a worked example and a link to the published definition it restates. Where an entry shows this result’s own value, it is that value set against a threshold the registry itself published. None of them is a statement about the variant you looked up.

What a result is measured against

Where the change sits, and the map the position is counted on. None of this says whether a change matters; it says which change is being talked about.

Genome build: The assembly of the human reference genome that a coordinate is counted against. GRCh38 is the current assembly and GRCh37 its predecessor. The two differ in sequence content and in the offsets that follow from it, so a chromosome and a position identify a base only once the assembly is named.rs334 is chr11:5,227,002 on GRCh38 and chr11:5,248,232 on GRCh37 — one base, two coordinates 21,230 apart.Genome Reference Consortium
Position: A chromosome and a 1-based offset along it: the first base of a chromosome is position 1. VCF and HGVS both count this way. SPDI does not — it is 0-based and interbase, so the same base is one lower there.chr11:5227002 on GRCh38 is the HBB base that carries the sickle-cell change; the SPDI for the same event begins NC_000011.10:5227001.Genome Reference Consortium
Gene: A named locus, given a unique symbol by the HGNC and a stable numeric ID by NCBI Gene. The symbol is the label used on this card; it can be reassigned, while the numeric ID cannot.HFE is NCBI Gene 3077 and still carries HLA-H among its aliases, so the literature on one gene sits under two symbols while the numeric ID is stable.NCBI Gene
dbSNP rsID: A Reference SNP cluster ID: dbSNP’s accession for a *locus and its alleles*, grouping every submitted record that maps to the same position with the same change. One rsID therefore covers every allele catalogued at that position, not one of them. A retired rsID is kept and points at the one it merged into.rs334 covers three alleles at chr11:5,227,002 — T>A, T>C and T>G. Only T>A is the sickle-cell allele, so "rs334" alone does not name a single change.dbSNP at NCBI
Reference sequence: The versioned accession a description is numbered against, written before the colon. A gene has many transcripts, and coding positions are counted from each transcript’s own start codon, so one genomic change takes a different c. number on each. The version suffix is part of the identity: a revision can renumber the same variant.NM_005957.5:c.665C>T and NM_001330358.2:c.788C>T are the same MTHFR variant, numbered on two transcripts. Both canonicalise to NC_000001.11:11796320:G:A.HGVS nomenclature: reference sequences
HGVS coding notation: A description numbered along a transcript’s coding sequence. c.1 is the A of the initiation codon; there is no c.0. Positions 5′ of the start codon are negative, positions 3′ of the stop codon carry an asterisk, and an intronic position is written as the nearest coding base plus or minus an offset.c.665C>T is a substitution 665 bases into the coding sequence. c.-1286T>A is 1,286 bases before the start codon. c.1521_1523del removes three bases. c.1234+1G>A is the first base of the following intron.HGVS nomenclature
HGVS protein notation: The predicted consequence numbered along the protein: the reference amino acid, its residue number, then the replacement. Three-letter and one-letter codes are both valid. "Ter" (or "*") is a stop codon and "fs" marks a frameshift. A p. description is a prediction from the sequence unless the source says the protein itself was observed.p.Glu7Val and p.E7V are the same change. p.Gln1756ProfsTer74 is a frameshift from residue 1756 that runs 74 residues before hitting a stop.HGVS nomenclature
Filed as: The exact description a registry indexes the record under. An insertion or deletion inside a repeat has several equally valid placements, and HGVS requires the 3′-most while VCF and SPDI conventions do not, so registries legitimately file one event at different coordinates. Both spellings are shown; equivalence is settled by canonical SPDI, never by position.BRCA1 c.5266dup is g.43057065dup at dbSNP and g.43057062_43057063dup at ClinVar — three bases apart, one variant.HGVS nomenclature

What the change does to the protein

Predictions worked out from where the change falls in the sequence. They describe a mechanism, not an outcome in a person.

Consequence: A Sequence Ontology term assigned by Ensembl’s Variant Effect Predictor from the position of the change relative to a transcript’s exons, reading frame and splice sites. It is a prediction from annotation alone: no transcript or protein was measured to produce it, and a variant has one term per transcript.The same base can be missense_variant on one transcript and intron_variant on another, because the transcripts differ in which exons they include.Ensembl consequence terms
Missense variant: A substitution inside a codon that changes which amino acid the codon specifies. Length and reading frame are unaffected; one residue differs.HBB codon 7, GAG → GTG: glutamate becomes valine. Written NM_000518.5:c.20A>T, p.Glu7Val — the sickle-cell allele.Ensembl consequence terms
Synonymous variant: A substitution inside a codon that leaves the amino acid unchanged, because the genetic code is degenerate — most amino acids have several codons. The protein sequence is identical; codon usage, splicing regulators and mRNA structure are not necessarily.GAA → GAG: both codons specify glutamate.Ensembl consequence terms
Frameshift variant: An insertion or deletion whose length is not a multiple of three. Translation reads codons in fixed groups of three from the start codon, so every codon downstream of the change is re-partitioned and the sequence from that point is unrelated to the original, usually terminating at a premature stop.Delete one base and every codon after it is new: … AAG CTG GAC ATC … Lys Leu Asp Ile … AAG TGG ACA TC … Lys Trp Thr … Delete three and the frame survives — that is an in-frame deletion instead.Ensembl consequence terms
Stop gained: A substitution that turns a sense codon into one of the three stop codons TAA, TAG or TGA, truncating the protein. Also called a nonsense variant. A premature stop before the last exon junction usually triggers nonsense- mediated decay, so the truncated protein is often not made at all.CGA → TGA: arginine becomes a stop, written p.Arg…Ter or p.R…*.Ensembl consequence terms
Stop lost: A substitution in the stop codon that makes it code for an amino acid, so translation continues into the 3′ untranslated region until the next in-frame stop, extending the protein.TGA → TGG: the stop becomes tryptophan and translation reads on.Ensembl consequence terms
Start lost: A change in the initiation codon, so translation cannot begin there. The next downstream AUG may be used instead, giving a shortened protein or none.ATG → ACG at c.1_3: the initiation codon is gone.Ensembl consequence terms
Splice donor variant: A change in the two bases at the 5′ end of an intron — the almost invariant GT of the donor site, at positions +1 and +2. The spliceosome recognises the intron by these bases, so a change here commonly causes exon skipping or use of a cryptic donor.c.1234+1G>A destroys the G of the donor GT.Ensembl consequence terms
Splice acceptor variant: A change in the two bases at the 3′ end of an intron — the almost invariant AG of the acceptor site, at positions −1 and −2 relative to the next exon.c.1235-1G>A destroys the G of the acceptor AG.Ensembl consequence terms
Splice region variant: A change within 1–3 bases of an exon boundary, or 3–8 bases into an intron: near the splice site but outside the invariant donor and acceptor pairs. Effects on splicing here are possible rather than expected.c.1234+5G>A is five bases into the intron — the splice region, not the donor.Ensembl consequence terms
In-frame deletion: A deletion whose length is a multiple of three, so whole codons are removed and the reading frame downstream is preserved. The protein is shorter by the residues deleted and otherwise unchanged in sequence.CFTR c.1521_1523del removes three bases and one residue: p.Phe508del, the commonest cystic-fibrosis allele.Ensembl consequence terms
In-frame insertion: An insertion whose length is a multiple of three, so whole codons are added and the reading frame downstream is preserved.Three bases inserted between two codons adds one residue: p.Ala…dup.Ensembl consequence terms
Intron variant: A change inside an intron and outside the splice donor, acceptor and splice- region windows. Introns are excised before translation, so the protein sequence is unaffected unless splicing or a regulatory element is.c.1234+150A>G sits 150 bases into an intron, well clear of either splice site.Ensembl consequence terms
5′ UTR variant: A change in the transcribed but untranslated region 5′ of the start codon. The protein sequence is unaffected; translation initiation, upstream open reading frames and transcript stability may not be.CFTR c.-1286T>A sits 1,286 bases before the start codon.Ensembl consequence terms
3′ UTR variant: A change in the transcribed but untranslated region 3′ of the stop codon, where microRNA binding sites and polyadenylation signals sit. The protein sequence is unaffected; transcript stability and abundance may not be.c.*84C>T is 84 bases past the stop codon.Ensembl consequence terms
Upstream gene variant: A change outside the transcript, 5′ of it, within the window the annotator reports on — 5,000 bases by default in Ensembl’s VEP. It is a statement about proximity, not about a promoter: VEP does not assert a regulatory role here.A change 2 kb before the transcription start site of the nearest gene.Ensembl consequence terms
Downstream gene variant: A change outside the transcript, 3′ of it, within the same default 5,000-base window. As with upstream, it reports proximity and nothing more.A change 3 kb past the end of the nearest transcript.Ensembl consequence terms
Non-coding transcript variant: A change in an exon of a transcript with no coding sequence — a long non-coding RNA, a pseudogene, or a retained intron. There is no reading frame, so no protein consequence is predicted.A change in an exon of XIST or MALAT1, neither of which codes for a protein.Ensembl consequence terms
Coding sequence variant: A fallback term: the change is within the coding sequence, but the effect on the protein could not be determined — typically because the alternate allele is unspecified or the local sequence is ambiguous.A record whose alternate allele is given as N falls here rather than under missense_variant or synonymous_variant.Ensembl consequence terms
MANE Select: Matched Annotation from NCBI and EMBL-EBI: one representative transcript per protein-coding gene, agreed between RefSeq and Ensembl, identical between them in the 5′ UTR, CDS and 3′ UTR, and aligned to GRCh38 without mismatch. It is a matched *pair* of accessions — one RefSeq, one Ensembl — which is what "matched annotation" means.MANE Select is a citation default, not a claim of biological primacy. A variant can be numbered on another transcript and both descriptions be correct, and the transcript most relevant to a particular variant may not be the MANE one. It is defined against GRCh38 only.HFE’s MANE Select is NM_000410.4 paired with ENST00000357618.10, protein NP_000401.1, out of 15 transcripts NCBI lists for the gene.MANE, at NCBI

What a model predicted

Scores from computer models, each trained on variants somebody had already classified. They are predictions rather than observations, and they disagree with each other routinely.

Computational predictions: Scores from models that estimate, from sequence and alignment features alone, whether a variant disrupts protein function or splicing. Each is a model output, not an observation, and most were trained on variants somebody had already classified — frequently on overlapping training sets.Their scales run in different directions and are not comparable to one another: SIFT calls ≤ 0.05 deleterious while REVEL’s 0.05 is unremarkable. Because they share training data, agreement between them is weaker evidence than the number of agreeing models suggests, and the ACMG/AMP framework treats them as supporting evidence at most.A variant scoring 0.03 at SIFT and 0.03 at REVEL is called deleterious by the first and unremarkable by the second — the same number on two opposite scales.gnomAD on the annotations it carries
REVEL: REVEL is an ensemble score for missense variants, combining thirteen other tools.Runs from 0 to 1, higher meaning more likely to be disease-causing in its authors’ training set. gnomAD reports the maximum across the transcripts it annotated.Ioannidis et al. 2016, the REVEL paper
CADD: CADD is how deleterious a substitution is predicted to be, across the whole genome.A scaled score: 10 means the variant is in the most deleterious 10% of possible substitutions by CADD’s own ranking, 20 the top 1%, 30 the top 0.1%.CADD, on how to read the score
SpliceAI: SpliceAI is how likely a variant is to change splicing.A delta score from 0 to 1. Its authors suggest 0.5 as a balanced cutoff, 0.2 as a high-recall one and 0.8 as high-precision. gnomAD reports the maximum of the four delta scores.Jaganathan et al. 2019, the SpliceAI paper
Pangolin: Pangolin is how likely a variant is to change splicing, by a second and independent model.A score from 0 to 1, higher meaning a larger predicted change in splice-site usage. gnomAD reports the largest of its delta scores.Zeng and Bhatt 2022, the Pangolin paper
phyloP: phyloP is whether a base has changed more or less often across species than drift alone predicts.Positive means slower change than expected — conservation; negative means faster — acceleration. It is a property of the position in an alignment of genomes, not of any one substitution at it.Pollard et al. 2010, the phyloP paper
SIFT: SIFT is whether an amino-acid substitution is tolerated at that position across homologues.Runs from 0 to 1, and is read the opposite way round from most of these: its authors call a score at or below 0.05 "deleterious". gnomAD reports the maximum across transcripts.SIFT, from its authors
PolyPhen-2: PolyPhen-2 is the predicted structural and evolutionary impact of an amino-acid substitution.Runs from 0 to 1, higher meaning more damaging by its model. Its authors report it alongside a qualitative call whose boundaries depend on which of their two training sets is used.PolyPhen-2, from its authors

What submitters concluded, in ClinVar

ClinVar’s own vocabulary. Each label is one submitter’s conclusion about a change and a condition on a date, and the review status says how much checking sits behind it.

ClinVar: An archive at NCBI of variant classifications submitted by clinical laboratories, expert panels, research groups and locus-specific databases. ClinVar stores and aggregates assertions; it does not generate them, does not re-interpret evidence, and does not adjudicate between submitters who disagree. A variant is in ClinVar only because somebody deposited a record.Variation 17677 (BRCA1 c.5266dup) carries 99 submissions from separate submitters, each with its own classification, condition and date.ClinVar’s definitions
Submission: One submitter’s assertion about one variant against one condition, with its own classification, review status, evidence and date of last evaluation. It is accessioned SCV and remains under the submitter’s control.Two labs asserting on the same variant produce two SCV records, which may classify it differently and are both retained.ClinVar identifiers
Variation ID: ClinVar’s integer identifier for a variant, stable across every submission about it and used as the Entrez uid for the record. Written as an accession it is VCV plus the same number zero-padded to nine digits.Variation ID 17677 is accession VCV000017677.ClinVar identifiers
SCV accession: A submitted record: one submitter, one variant, one condition. The submitter owns it and is the only party that can revise it. Every classification the card shows in a submission row is an SCV.The suffix after the dot is the record’s version, incremented by the submitter on each revision.The suffix is the submitter’s own revision count, so a record ending .5 has been revised four times since it was first deposited.ClinVar identifiers
RCV accession: ClinVar’s aggregation of every submission about one variant *against one condition*. A variant asserted on for three conditions has three RCV records, each with its own aggregate classification.The suffix after the dot increments whenever a submission beneath it is added, revised or withdrawn.MTHFR variation 3520 has 10 RCV records — one per condition it has been asserted against — over 23 submissions.ClinVar identifiers
VCV accession: ClinVar’s aggregation of every submission about one variant across all conditions. There is exactly one per variant, and its number is the Variation ID padded to nine digits.The suffix after the dot increments whenever a submission beneath it is added, revised or withdrawn.VCV000017677 aggregates all 99 submissions on BRCA1 c.5266dup, across every condition any of them named.ClinVar identifiers
Condition: The disease or phenotype a submitter assessed the variant against, reported by the submitter and mapped by ClinVar to a MedGen concept. A classification is only meaningful with its condition attached: the same variant can be pathogenic for one and benign for another.A submitter naming no trait but supplying the cross-reference OMIM:601634 is resolved through ClinVar’s TraitMapping to a MedGen preferred name.ClinVar’s definitions
Pathogenic: The submitter judged the evidence sufficient to conclude the variant causes the condition named beside it. Under the ACMG/AMP framework most submitters use, this is a threshold on combined evidence criteria, asserted on a date — not a measurement, and not a statement about a particular carrier.On a three-axis record the germline classification can be Pathogenic while the somatic and oncogenicity axes say something else; each is shown separately.ClinVar’s definitions
Likely pathogenic: The same conclusion at a lower evidence threshold. The ACMG/AMP guideline defines "likely" as greater than 90% certainty, and says explicitly that the figure is for conceptual clarity rather than a computed probability.A variant can move from Likely pathogenic to Pathogenic, or out of both, as segregation or functional evidence accumulates.ClinVar’s definitions
Uncertain significance: The evidence is insufficient, or conflicting within one submitter’s own assessment, to place the variant in either direction. It is an absence of evidence, not an intermediate degree of pathogenicity, and it is the largest class in the archive.A VUS reclassifies in either direction as evidence arrives; the date of last evaluation on each row is what says how current the assessment is.ClinVar’s definitions
Likely benign: The submitter judged the variant not to cause the condition, at the same >90% threshold the ACMG/AMP guideline sets for "likely".ClinVar aggregates Benign and Likely benign as agreeing, so a record with both is not counted as conflicting.ClinVar’s definitions
Benign: The submitter judged the variant not to cause the condition named beside it. It is a conclusion scoped to that condition and that submitter’s evidence, not a general statement that the variant has no effect.A variant can be Benign for one condition and carry a separate assertion of drug response for another.ClinVar’s definitions
Conflicting classifications: ClinVar’s aggregate value when submissions with assertion criteria disagree across the benign/pathogenic divide. Benign and Likely benign count as agreeing, as do Pathogenic and Likely pathogenic; disagreement between those groups produces this. ClinVar reports the disagreement rather than resolving it.BRAF c.1799T>A has a germline aggregate of Conflicting classifications of pathogenicity resting on 6 of its 22 germline submissions.ClinVar’s definitions
Drug response: A germline classification term for a variant asserted for its effect on response to a drug — dosing, efficacy or adverse reaction — rather than for causing disease. It sits in the same field as Pathogenic and Benign and is not a point on the same scale.MTHFR variation 3520’s only germline aggregate is drug response, at three stars, while its 23 submissions span nine different classification terms.ClinVar’s definitions
Risk factor: The variant was asserted as contributing to risk of the condition rather than being sufficient to cause it — the term used for common alleles of modest effect and for modifiers.An allele raising risk a few percent in a population is filed here, not as Pathogenic.ClinVar’s definitions
Association: The variant was reported as statistically associated with the condition in a study population. Association is a property of a group comparison and does not establish causation or apply to an individual.A genome-wide association finding deposited as a ClinVar assertion carries this term.ClinVar’s definitions
Protective: The variant was asserted as reducing risk of the condition relative to the reference allele.It is the mirror of Risk factor: a direction of effect away from the condition, not a claim that the variant is harmless.ClinVar’s definitions
Affects: The variant was asserted to affect a phenotype that is not treated as a disease — a trait, or a measurable laboratory characteristic.A variant determining a blood group or an eye colour is filed here.ClinVar’s definitions
Not provided: The submitter deposited a record without an interpretation. The record is retained because it may still carry a condition, literature citations or observation counts.A submission of Not provided contributes nothing to ClinVar’s aggregate but still appears in the submission list and in its tally.ClinVar’s definitions
Review status: ClinVar’s statement of the evidence level behind a classification: whether assertion criteria were supplied, how many submitters supplied them, whether they agreed, and whether an expert panel or practice guideline is involved. It is rendered as zero to four stars and is the basis of ClinVar’s aggregation precedence, which is a hierarchy and not a vote.Four stars, practice guideline. Three, expert panel. Two, multiple submitters with criteria and no conflicts. One, a single submitter with criteria, or submitters with criteria who conflict. Zero, no criteria provided.ClinVar’s germline aggregate on BRCA1 variation 17677 rests on 1 of its 99 submissions — an expert-panel assertion, over 74 that provided criteria.ClinVar review status
Practice guideline: Four stars, ClinVar’s highest level: the classification is drawn from a published clinical practice guideline. It becomes the aggregate whatever else was submitted.It outranks everything below it, so a single guideline row settles the aggregate however many laboratories have submitted.ClinVar review status
Reviewed by expert panel: Three stars: a panel approved by ClinGen for that gene or disease reviewed the evidence and issued a single classification, which becomes the aggregate regardless of the other submissions.ENIGMA — the Evidence-based Network for the Interpretation of Germline Mutant Alleles — submits BRCA1 and BRCA2 assertions at this level.ClinVar review status
Multiple submitters, no conflicts: Two stars: more than one submitter provided assertion criteria and their classifications agree, treating Benign with Likely benign and Pathogenic with Likely pathogenic as agreement.Its full ClinVar string is "criteria provided, multiple submitters, no conflicts" — which contains the word "conflicts" and must be matched before the one-star conflicting status, or agreement reads as dispute.ClinVar review status
Conflicting classifications: One star: submitters provided assertion criteria and disagreed across the benign/pathogenic divide. The aggregate records the disagreement.Its ClinVar string is "criteria provided, conflicting classifications" — one star, against the two stars of the no-conflicts status.ClinVar review status
Single submitter, criteria provided: One star: exactly one submitter, who stated the assertion criteria used. No independent submission exists to agree or disagree.A single laboratory’s assertion with criteria sits here until a second arrives.ClinVar review status
No assertion criteria provided: Zero stars: the submitter did not state the criteria used. ClinVar aggregates such submissions only when no criteria-provided submission exists.On MTHFR variation 3520, 11 criteria-provided rows are outranked by one three-star assertion, and the zero-star rows are outranked by both.ClinVar review status

How common the change is

Counts from a large collection of people. How often something turns up is evidence about the change; it is not a conclusion about it.

Allele frequency: AC/AN: the count of chromosomes observed carrying the alternate allele, over the count of chromosomes successfully genotyped at that site. The denominator is chromosomes, not people, so on an autosome it is close to twice the number of individuals — and it varies from site to site with coverage.A high frequency in a large reference cohort is evidence against a variant being a highly penetrant cause of a rare severe disease, and the ACMG/AMP framework has a stand-alone benign criterion for it. Rarity carries no converse implication: almost every variant is rare.gnomAD Help/FAQ
gnomAD: The Genome Aggregation Database: a joint call set of exomes and genomes aggregated from disease-specific and population studies. Cohorts recruited for severe paediatric disease are removed, but gnomAD states that it holds no comprehensive phenotype metadata and that many samples now come from large biobanks which do include people with disease. It is a reference for allele frequency, not a healthy control cohort.AN — the chromosomes genotyped at a site — is the number that says how much the estimate rests on, and it differs by orders of magnitude between releases and between the exome and genome call sets.MTHFR rs1801133 has AN 1,461,758 in the v4 exomes and 251,468 in v2 — the same variant, estimated on nearly six times as many chromosomes.gnomAD
gnomAD genomes: The whole-genome call set: fewer individuals than the exomes, but with coverage outside coding exons — the only one of the two that reports intergenic, deep-intronic and regulatory positions.A 5′ UTR or deep-intronic variant has a genome frequency and no exome frequency, because exome capture does not target it.gnomAD on dataset selection
gnomAD exomes: The exome-capture call set: an order of magnitude more individuals than the genomes, covering the targeted coding regions only, with coverage falling off at the edges of capture baits.A coding variant usually carries both an exome and a genome frequency; the exome AN is the larger, so its estimate is the more precise.gnomAD on dataset selection
GroupMax filtering allele frequency (95%): The lower bound of the 95% confidence interval around the allele frequency, computed within each genetic ancestry group, and reported for the group whose FAF is highest — not the group whose raw AF is highest. gnomAD excludes bottlenecked groups from the calculation: Amish, Ashkenazi Jewish, European Finnish and Remaining individuals, and Middle Eastern in the genomes.Whiffin et al. 2017 defines it as the maximum credible population AF: if the FAF exceeds the threshold set for a specific disease, from its prevalence, penetrance and allelic heterogeneity, the observed count is not compatible with pathogenicity for that disease. The threshold is per-disease and is the clinician’s to set — gnomAD does not supply one and neither does this tool.gnomAD on filtering allele frequency
Genetic ancestry group: A label assigned by a random-forest classifier trained on samples of known ancestry, using the first 20 principal components of genotype data as features. Samples not assigned above the per-group probability threshold are placed in "Remaining individuals". The grouping is derived from genotypes, not from self-report, and is not a race, nationality or ethnicity.The assignment thresholds differ by group — 0.96 for East Asian and South Asian, 0.56 for Middle Eastern — so the groups are not equally well resolved, and their sizes differ by orders of magnitude. A frequency within a group describes the samples gnomAD assembled, which are not a random sample of any population.gnomAD v4 reports afr, ami, amr, asj, eas, fin, mid, nfe, sas and remaining. Sub-group clustering was not performed for v4.gnomAD on genetic ancestry groups
Homozygous: nhomalt: the number of individuals in the callset called homozygous for the alternate allele. A count of individuals, unlike AC and AN, which count chromosomes.For a variant proposed as causal for a recessive condition, homozygotes among people not ascertained for that condition are evidence against it. gnomAD is not a healthy cohort — it excludes severe paediatric disease but not adult disease — so the count constrains an argument rather than settling it.gnomAD on genetic ancestry groups
Quality-control filter: A flag from gnomAD’s variant QC saying the site failed one of its checks. AS_VQSR is the allele-specific variant quality score recalibration; AC0 means no sample had a high-quality genotype (GQ ≥ 20, DP ≥ 10, and allele balance > 0.2 for heterozygotes); InbreedingCoeff means an excess of heterozygotes against Hardy–Weinberg expectation, below −0.3.A flag is a statement about the sequencing evidence for the call, not about the biological effect of the variant. A frequency carrying a filter is less well supported than the same number without one.A variant flagged AC0 has an allele count derived entirely from genotypes that failed gnomAD’s own quality thresholds.gnomAD on variant quality control

How the gene behaves as a whole

Population statistics about the gene the change falls in. They say nothing about any one change inside it.

Gene constraint: A measure of depletion of variation in a gene against a neutral expectation. gnomAD fits a mutational model that predicts the number of variants expected per transcript from local sequence context and CpG methylation, then compares that with the observed count — unique SNVs at alternate allele frequency < 0.1% passing variant QC. Significant depletion is read as negative selection.It is a property of the gene, computed on the MANE Select transcript in v4, and says nothing about any particular variant within it. gnomAD flags 1,348 MANE Select transcripts as outliers and excludes them, leaving 17,063.A gene can be highly constrained and still carry common harmless variants, and a gene under no constraint can carry a variant that causes disease. The metric ranks genes, not the variants inside one.gnomAD on gene constraint
LOEUF: Loss-of-function observed/expected upper bound fraction: a conservative upper bound on the ratio of observed to expected predicted loss-of-function variants in a gene. gnomAD v4 computes it in a Bayesian frame — a Gamma posterior on the true underlying pLoF count under a non-informative prior, taking the 95th percentile and dividing by the neutral expectation. The original formulation was the upper bound of a central 90% Poisson confidence interval; the interpretation is unchanged, the quantity is not.Low means constrained. gnomAD recommends using it as a continuous value, and suggests < 0.45 — the 15th percentile of 17,063 MANE Select transcripts — for Mendelian interpretation where a threshold is needed. Thresholds do not carry over between releases: values rose from v2 to v4 as sample size grew, so a v2 cut-off selects a different set of genes when applied to v4.gnomAD on gene constraint
pLI: The probability that a gene belongs to the haploinsufficient class. An expectation-maximisation algorithm assigns each transcript a probability of belonging to each of three classes — null, recessive and haploinsufficient — calibrated on the empirical loss-of-function o/e of known recessive genes (0.706) and known severe haploinsufficient genes (0.207). pLI is the third of those probabilities.High means intolerant, which is the opposite direction to LOEUF and o/e. pLI > 0.9 is the widely used cut-off. It is close to dichotomous: gnomAD notes that an intermediate value between 0.1 and 0.9 usually means the gene was too small to categorise confidently, rather than that it is partly intolerant.gnomAD on gene constraint
o/e LoF: The point estimate of observed over expected predicted loss-of-function variants — the ratio LOEUF puts an upper bound on. It is a direct proportion: 0.4 means 40% of the expected pLoF variants were observed.gnomAD says it is essential to read o/e with its 90% confidence interval, because precision depends on gene size and sample size; that is what LOEUF exists to supply. A short gene can show a low o/e on very few expected variants and not be constrained at all.gnomAD on gene constraint
Missense z: A signed z-score for the deviation of the observed missense count from the count expected under the same mutational model, over rare variants (MAF < 0.1%). gnomAD takes the square root of the chi-squared deviation, signs it positive when fewer variants were observed than expected, and rescales it by the standard deviation of a mirrored distribution of the scores.Positive means fewer missense variants than expected — increased constraint. Negative means more than expected. gnomAD publishes no threshold for it and recommends LOEUF as the metric to lead with.gnomAD on gene constraint

What has been associated with it

Patterns reported between a place in the genome and a measured trait, across groups of people. They describe a study population, not a person, and an association is not a cause.

Trait association: A reported statistical association between a variant and a trait in a genome-wide association study: carriers of one allele differed in the trait more than chance would account for, across a study population. Association is a property of that comparison. It does not establish that the variant is causal — the tested variant is usually a tag in linkage disequilibrium with whatever is — and it says nothing about an individual.The catalogue records the association, the study and the paper; it does not record whether the finding later replicated.GWAS Catalog documentation
GWAS Catalog: A curated database of published genome-wide association findings, maintained by EMBL-EBI and NHGRI. Curators extract associations meeting the catalogue’s inclusion criteria from the literature and map the reported traits onto Experimental Factor Ontology terms so that studies can be compared.It records what was published, per study. It does not weigh replication, and a variant with many entries has been studied often, which is not the same as being well established.EFO traits are measurement names rather than clinical wording: rs1801133’s strongest are filed as "homocysteine measurement", not "hyperhomocysteinemia".GWAS Catalog documentation
p-value: The probability of observing an association at least this strong if the null hypothesis of no association were true. It is a statement about data under an assumption, not the probability that the association is real.Genome-wide studies test roughly a million independent common variants, so the conventional significance threshold is 5×10⁻⁸ — 0.05 Bonferroni-corrected for that many tests — rather than the 0.05 used for a single test. A p-value bounds the role of chance and says nothing about the size of the effect: with a large enough sample, a negligible effect reaches any threshold.GWAS Catalog documentation
Effect size: The estimated magnitude of the association per copy of the reported allele: a beta, in the trait’s own units, for a continuous trait, or an odds ratio for a binary one. This is the quantity the p-value says nothing about.Effect sizes for common variants are typically small — odds ratios near 1.1 are usual for complex traits — and they are estimated in a specific study population, so they do not transfer unchanged to another.GWAS Catalog documentation
Reported allele: The allele the study measured the effect against. Direction is relative to this allele, so an identical finding reported against the other allele carries the inverse effect size — an odds ratio of 1.4 becomes 0.71.The catalogue writes a bare "-" where a study reported no effect size at all, which is an absence rather than a negative value.GWAS Catalog documentation

Published work

What has been written about a change, and how it is counted.

Europe PMC: A literature database maintained by EMBL-EBI. It contains everything in PubMed and additionally indexes the full text of the open-access subset, along with preprints, patents, theses and agricultural records — which is why a query can match a variant named only in a supplementary table.Searching one variant’s identifiers across Europe PMC and PubMed on the same boolean: MTHFR rs1801133 returns 2,878 records against 897, and every one of the PubMed-only records is also in Europe PMC.About Europe PMC
Cited by: The number of records in Europe PMC’s own citation network that cite this one. The network is built from the reference lists Europe PMC can parse, so it is not a complete count of citations in the literature and differs from the counts Scopus or Web of Science report.It measures attention, not correctness, and it accumulates with age, so it systematically favours older papers over recent ones on the same subject.About Europe PMC
PubMed: NLM’s index of biomedical literature. It holds bibliographic metadata — title, abstract, authors, journal — and MeSH indexing, and does not hold full text. A query therefore matches title, abstract and subject headings only.PubMed expands query terms onto MeSH headings automatically, which is its recall mechanism and also why a result can contain no literal occurrence of the term searched for. Punctuated identifiers are split by that expansion.Quoting does not defend against it: "APOE:c.388T>C" is translated to "apolipoproteins e"[MeSH] AND "c"[All Fields] — the gene ANDed with a letter.PubMed
Most relevant: PubMed’s Best Match ordering, which is what sort=relevance selects. PubMed first weights each record by how many of the query terms it matches and in which fields, then re-ranks the top of that list with a machine-learning model trained on what users of PubMed actually clicked. Recently published articles are weighted up.A sample drawn this way is the top of a ranking, not a random sample of the matching records, and the ranking is tilted towards recent and towards frequently-clicked papers. Proportions measured over it describe those papers; they do not estimate the proportions in the whole result set.On the same BRCA1 query, sort=relevance and PubMed’s default ordering return entirely different first pages of the same 348 records.PubMed’s help on Best Match
Subject heading: A Medical Subject Headings descriptor assigned by an NLM indexer to say what a paper is about, marked with an asterisk when it is a major topic of the paper rather than incidental. MeSH is a controlled vocabulary arranged in a tree, and a descriptor’s tree numbers are what identify it as a disease (category C) or a mental disorder (F03) rather than a gene, a chemical or a method.A heading records what a paper is indexed as being about. It carries no information about what the paper concluded, or whether the conclusion held up.Stroke sits at both C10.228.140.300.775 and C14.907.253.855, so a descriptor has to be checked at every position it occupies, not the first.NLM's MeSH documentation

The evidence codes a rule can decide

Named categories of evidence from the ACMG/AMP guideline. Each is one argument about a change, weighed and then combined with the rest by a curator; a code on its own is evidence, and the classification is the combination, which VarInsight does not perform.

ACMG evidence codes: Twenty-eight named criteria from the 2015 ACMG/AMP guideline, each a category of evidence that a sequence variant does or does not cause disease. Sixteen point towards pathogenic and twelve towards benign, and each carries a weight: stand-alone, very strong, strong, moderate or supporting. A curator applies the criteria that the available evidence supports and then combines them by a published table into one of five classifications.A code is one piece of evidence, not a conclusion. The classification comes from the combining rules applied to every criterion a curator could evaluate, so a subset of the codes does not determine it and cannot be read as pointing at it.PM2 is the criterion for absence from population databases. It is applied at supporting weight since ClinGen revised it in 2020, which is why it is written PM2_Supporting.Richards et al. 2015, the ACMG/AMP standards and guidelines
PVS1: The very strong pathogenic criterion for a null variant — nonsense, frameshift, a change at a canonical splice site one or two bases into the intron, a lost initiation codon, or a deletion of one or more whole exons — in a gene where loss of function is an established mechanism of disease.The consequence class is only the entry condition. ClinGen’s 2018 recommendation adds a decision tree covering nonsense-mediated decay, how much of the protein is removed, whether the exon can be skipped and whether a downstream start codon rescues it, and the gene’s disease mechanism has to be a curated statement rather than a constraint metric.A nonsense change in the last exon escapes nonsense-mediated decay, so the transcript is made and a shortened protein is produced. The decision tree drops PVS1 to a lower weight there, or withdraws it, depending on how much of the protein is lost.Abou Tayoun et al. 2018, the ClinGen SVI PVS1 recommendations
BA1: The stand-alone benign criterion: an allele frequency above 5% in a general continental population dataset of at least 2,000 called alleles, in a gene with no gene-specific or variant-specific modification of the rule. ClinGen names the datasets as African, East Asian, European (Non-Finnish), Latino and South Asian, and excludes the Finnish as a founder population.The exception list is part of the rule rather than a footnote to it. ClinGen excepts nine named variants for which evidence of pathogenicity exists despite a high frequency, and a variant on that list does not meet BA1 however common it is.HBB c.20A>T, which causes sickle cell disease, is above 5% in gnomAD’s African/African American exomes, because carrying one copy protects against malaria. A frequency rule cannot see that, which is why a code is not a classification.Ghosh et al. 2018, ClinGen’s updated BA1 recommendation
PM2_Supporting: The criterion for a variant absent from population databases, or at extremely low frequency where the disorder is recessive. ClinGen reduced it from moderate to supporting weight in September 2020, on the grounds that rarity does not reach the odds of pathogenicity a moderate criterion assumes — which is why the weight is written into the code.Absence is decidable from a database. What counts as extremely low is not: it follows from the disorder’s prevalence, penetrance and genetic heterogeneity, and expert panels set it per gene.A variant gnomAD holds no record of is absent from it. A variant gnomAD has seen in 101 of 1.46 million alleles is not absent, and whether it is rare enough depends on the disorder being considered.ClinGen SVI recommendation for PM2, version 1.0 (September 2020)
PP3: The pathogenic criterion for computational evidence of a deleterious effect. ClinGen calibrated it in 2022 against the odds of pathogenicity a criterion of each weight assumes, so a predictor score maps to a weight rather than to a yes or no: for REVEL, 0.644 and above is supporting, 0.773 and above moderate, 0.932 and above strong.The calibration asks that one tool be chosen genome-wide, before the score is seen, because picking the tool that gives the answer wanted is a multiple-testing problem. It also asks that PP3 and PM1 together not exceed strong weight.A REVEL score of 0.842 falls between 0.773 and 0.932, so the calibration places it at moderate weight rather than the supporting weight the 2015 guideline gave every computational argument.Pejaver et al. 2022, ClinGen’s PP3/BP4 calibration
BP4: The benign counterpart of PP3: computational evidence suggesting no impact. On the same 2022 calibration a REVEL score of 0.290 or below is supporting, 0.183 or below moderate, 0.016 or below strong and 0.003 or below very strong. For a change that alters no amino acid, the evidence read instead is a splicing prediction.The band between the benign ceiling and the pathogenic floor — 0.290 to 0.644 for REVEL — supports neither code. That gap is a result of the calibration rather than a hole in it: the scores in it do not shift the odds enough to count as evidence either way.A REVEL score of 0.535 is in that band, so neither PP3 nor BP4 applies and the computational argument contributes nothing in either direction.Pejaver et al. 2022, ClinGen’s PP3/BP4 calibration
BP7: The benign supporting criterion for a synonymous change — one that leaves the amino acid sequence unaltered — where splicing prediction shows no effect on the splice consensus and no new splice site created. ClinGen reads that as SpliceAI’s maximum delta score at or below 0.1, and extends the criterion to intronic changes far enough from the exon boundary.The 2015 wording also asked that the nucleotide not be highly conserved. ClinGen dropped that condition in 2023, having measured that adding a conservation filter gave no improvement in the ability to identify changes that do not disrupt splicing.A synonymous change with a SpliceAI maximum delta score of 0.00 meets the criterion even where the base is highly conserved, because conservation at a position is a fact about the alignment rather than about splicing.Walker et al. 2023, the ClinGen SVI splicing recommendations
PM4: The moderate pathogenic criterion for a change in the length of the protein: an in-frame insertion or deletion outside a repeat region, or a stop-loss change, where the termination codon is lost and translation continues past the normal end of the protein.The two limbs carry different conditions. The stop-loss limb has none beyond the consequence. The in-frame limb applies only outside a repetitive region, which takes a repeat annotation to establish.A three-base deletion removes one amino acid and leaves the reading frame intact, so the rest of the protein is translated normally. That is the in-frame limb, and whether the three bases fall inside a repeat is what decides it.Richards et al. 2015, the ACMG/AMP standards and guidelines

What these definitions are not

They define words, not results. Nothing here is a conclusion about a variant you looked up, and nothing here is medical advice. Where a definition gives a threshold — pLI at 0.9, LOEUF at 0.45 — that threshold is the registry’s own published suggestion for reading its own number, linked beside it, and not a rule this tool invented. Registries revise them: gnomAD’s LOEUF suggestion is now < 0.45, and its own documentation warns that a threshold set on an earlier release selects a different set of genes when applied to a later one. On a result card each of these entries also shows where that result’s own value falls against the threshold; this page has no result, so it shows the definitions alone. If you are reading your own genetic data, take it to a licensed clinician or a certified genetic counsellor.