Reading a Variant's Evidence Card
7 min read · Updated October 30, 2026
A variant lookup returns a clinical classification, population frequency and a handful of predictor scores. Read in the wrong order they contradict each other; read in the right order one of them usually settles the question.
A single worked lookup — rs1042522, the common TP53 codon 72 polymorphism — contains four separate traps, all of them visible in the output.
An rsID is not a variant
Looking up rs1042522 on GRCh38 returns four records, all at chr17:7,676,154: G>A, G>C, G>T and a reference record. ClinVar classifies them differently — Conflicting classifications of pathogenicity for G>A, Benign reviewed by an expert panel for both G>C and G>T, and Benign from a single submitter for the reference record.
An rsID names a position with known variation, and a multi-allelic site has several alleles at it. Picking the first record, or letting a tool pick one silently, is how a report ends up describing a different change from the one that was sequenced. Give the allele, not just the rsID, whenever you have it.
Read the frequency before the label
The G>C record — the one that produces the familiar Pro72Arg change — has a gnomAD exome allele frequency of 0.668 and a genome frequency of 0.627, ranging from 0.38 in African-ancestry samples to 0.73 in Finnish and 0.74 in non-Finnish European ones.
An allele carried by two thirds of people is not a Mendelian disease allele. The frequency alone settles it, and it settles it faster than any amount of reading about the gene.
This is the single most useful habit in variant interpretation: look at the population frequency first, because a common allele cannot be a highly penetrant cause of a rare disease, whatever any individual submission says.
One record is not ClinVar's classification
Underneath that allele's classification sit ten per-condition records: nine Benign and one Pathogenic. Several of the Benign ones are marked criteria provided, multiple submitters, no conflicts, for conditions including Li-Fraumeni syndrome and hereditary cancer predisposition; one is filed under TP53 POLYMORPHISM. The single Pathogenic record, for lip and oral cavity carcinoma, carries ClinVar's weakest review status: no assertion criteria provided.
Any lookup that headlines the most severe of those records shows this allele as Pathogenic. ClinVar itself does not: its classification of the variant is Benign, reviewed by an expert panel. That is not a vote over the records — when a practice guideline or an expert panel has classified a variant, ClinVar uses that classification and its review status for the whole record, whatever the individual submissions say.
So read the classification and its stars before any single record. Four stars is a practice guideline, three an expert panel, two several submitters with criteria who agree, one a single submitter or submitters who conflict, and none a record submitted without criteria. The records are still worth reading, for which conditions were considered — a classification aggregated across conditions is not a statement about the one you are asking about.
Predictor scores are per transcript, and they disagree
For the G>A record at the same position, SIFT returns eight predictions — seven Tolerated and one Deleterious, with scores from 0.172 down to 0.046. AlphaMissense returns thirteen, all likely benign, from 0.079 to 0.094.
The spread is not disagreement between tools; it is one tool scoring the change against eight different transcripts of the same gene, in which the affected residue sits in a different context. Quoting SIFT said deleterious from that output is quoting one transcript out of eight.
Nor are the scales aligned: SIFT is a probability where LOW means damaging, while AlphaMissense is a 0–1 score where HIGH means pathogenic — likely benign below 0.34, likely pathogenic above 0.564. The same number means opposite things depending on which column it is in.
Empty is a statement about the snapshot
Three of the four records return no gnomAD exome frequency at all. That means the frozen gnomAD release behind the lookup has no entry for that allele — not that the allele is absent from the population.
Know which parts are live and which are snapshots with dates. ClinVar's classification above is read live; the per-condition records under it come from a periodic snapshot and can lag it; gnomAD exomes are v2.1.1 from 2019 and the GRCh38 genomes an older v3 release. A frequency from the current gnomAD release will differ — and on this lookup the exome and genome columns are already two different releases.
For a clinical question, a card like this is a triage view, and the current databases are the record.
Frequently asked questions
Why does one rsID return several variants?
Because an rsID names a position where variation is known, and a multi-allelic position has more than one alternate allele. rs1042522 returns four records at the same coordinate, which ClinVar classifies three different ways. Specify the alternate allele rather than letting a tool choose one.
One ClinVar record says pathogenic but the allele frequency is high. Which is right?
Usually the frequency, and ClinVar's own classification will often say so too. A worked case: an allele with a gnomAD frequency of 0.67 has ten ClinVar records, nine Benign with criteria provided and one Pathogenic with no assertion criteria provided — and ClinVar classifies the variant Benign, reviewed by an expert panel. Read the classification and its stars, not the most severe record.
Why do SIFT and AlphaMissense give several scores for one variant?
One score per transcript. A substitution lands at a different position, and in a different protein context, in each transcript of the gene, so eight transcripts give eight scores and they need not agree — in a worked example, seven Tolerated and one Deleterious. Pick the transcript you care about rather than the most severe score.
What do the predictor scores actually mean?
Different things, on different scales. SIFT is low-is-damaging; AlphaMissense is high-is-pathogenic, with likely benign below 0.34 and likely pathogenic above 0.564. So two scores of the same magnitude can point in opposite directions. Neither is evidence of pathogenicity on its own under current interpretation guidelines; they are supporting evidence at best, and they are correlated with each other rather than independent.
Why is there no frequency for my variant?
Because the frozen release behind the lookup has no entry for that allele. That is a statement about the snapshot, not about the population — rare alleles are often absent from an older, smaller cohort. Check the current gnomAD release before concluding a variant is novel.
Related tools
Look up a variant by rsID, chrom:pos:ref:alt, or HGVS and get its ClinVar significance, gnomAD allele frequencies, and SIFT and AlphaMissense scores in one card.
Convert HGVS between c., g. and p.: a c. variant to genomic coordinates and its protein effect, or a genomic position back onto its MANE transcript — via a live Ensembl exon map.
Align a query to a reference and list substitutions, insertions and deletions with effects.
Search a gene and get its exon/CDS structure, a druggability/disease/drug/trial/paper dossier, and a GTEx + Human Protein Atlas expression fingerprint — all in one page.
Convert gene symbols, Ensembl, Entrez GeneID, RefSeq and UniProt accessions in batch, and find cross-species orthologs.