Module 03 · 70 min

Variant Interpretation: ACMG, VUS, and Penetrance

From a VCF line to a clinical recommendation — the ACMG/AMP framework and where it breaks.

FoundationsClinicalResearch

Core topics

What's covered

T01ACMG/AMP 2015 framework + 2023–2025 ClinGen specifications (PVS1, PS3, PP3 recalibration)
T02Bayesian point system (Tavtigian 2018) underlying the categories
T03Reading a ClinVar entry: submitter, criteria, conflicts, expert panel review
T04The VUS problem: drift, reclassification, family segregation
T05Penetrance vs pathogenicity — why 'pathogenic' ≠ 'will get sick'
T06Population frequency in gnomAD v4 and ancestry caveats

Learning objectives

By the end of this module you will be able to

L01Apply the ACMG criteria to classify a variant given evidence lines, including the recalibrated PP3/BP4 (Pejaver 2022).
L02Explain to a family why a 'pathogenic' BRCA1 variant does not mean 100% lifetime risk.
L03Recognize when a VUS is unlikely to be reclassified vs. when active follow-up is warranted.

Key takeaways

What you should walk away believing

Pathogenicity is a property of a variant; risk is a property of a person — penetrance bridges them and is often <50% for adult-onset conditions identified through population screening.
REVEL ≥ 0.932 now counts as PP3_Strong under the 2022 ClinGen recalibration — older labs may not yet apply this.
ClinVar conflicts are common; always inspect submitter quality and criteria before accepting a classification.
Penetrance estimates from clinically ascertained families are inflated relative to population-based estimates (UK Biobank, All of Us) by 2–5×.

Lesson · Foundations emphasis

What this means at your level

Foundations

Every genome carries millions of variants. Almost all are benign. The ACMG/AMP framework is a structured set of evidence rules used to classify a variant from Benign through Likely Benign, Variant of Uncertain Significance (VUS), Likely Pathogenic, to Pathogenic. The same variant can be classified differently by different labs, which is why ClinVar — a public database of submissions — matters.

Clinician deep-dive

When you receive a report, do three things: (1) confirm the variant is in ClinVar and check for conflicting interpretations; (2) check gnomAD frequency by ancestry — anything >1% in any ancestry is essentially never pathogenic for a rare disease; (3) ask whether penetrance estimates were derived from ascertained families (likely overestimated) or unbiased biobanks. For VUS, set a 1–2 year reanalysis trigger and counsel the family that classification may change.

Research note

The 2022–2024 ClinGen recalibration of PP3/BP4 (Pejaver et al.) made in-silico predictors quantitative rather than binary, with REVEL, AlphaMissense, and ESM1b as the strongest performers for missense. Splice prediction has moved from MaxEntScan/SpliceAI v1 to SpliceAI + Pangolin ensembling. ACMG SF v3.2 (2023) lists 81 genes for opportunistic secondary-finding return.

Myth-buster

A pathogenic variant means the person will get the disease.

Reality

Penetrance varies widely. BRCA2 lifetime breast-cancer risk is ~70% in clinically-ascertained families but ~20–30% in population biobanks. LDLR pathogenic variants for FH have ~40% penetrance for early MI in unselected populations. Counseling that conflates pathogenicity with certainty causes real harm.

Evidence-graded claims

What the data say

The ACMG/AMP 2015 criteria are the standard for germline variant classification

Universally adopted; ClinGen continues to refine gene/disease-specific specifications.

Audit grade A in this module →

REVEL outperforms older missense predictors at clinically relevant thresholds

ClinGen Sequence Variant Interpretation working group recalibration confirms it on independent test sets.

Audit grade A in this module →

AlphaMissense should be used as primary clinical evidence today

Strong benchmark performance, but ClinGen has not yet defined formal evidence-strength tiers; treat as supporting.

Audit grade C in this module →

Population biobank-based penetrance estimates are systematically lower than family-based estimates

Documented across BRCA1/2, LDLR, MYH7, MYBPC3, LMNA, and ATTR.

Audit grade A in this module →

Direct-to-consumer 'health risk' reports based on a few SNPs accurately predict individual disease risk

23andMe-style reports explain a small fraction of variance; PRS-based reports are better but still calibration-poor outside the training ancestry.

Audit grade E in this module →

See all claims for this module in the Evidence Grader →

Quick check

Test yourself

Q1A variant is reported as 'Likely Pathogenic.' Under the Tavtigian Bayesian formulation, this corresponds to roughly:

Q2Best initial action for a newly issued VUS in a clinically meaningful gene:

Glossary

Key terms & abbreviations

ACMG/AMP: American College of Medical Genetics & Association for Molecular Pathology — author of the dominant variant classification framework.
VUSVariant of Uncertain Significance: A variant with insufficient evidence to classify as benign or pathogenic. Approximately 30–50% of clinically reported rare variants.
Penetrance: Proportion of carriers of a given genotype who manifest the associated phenotype by a given age.
gnomAD: Aggregated population variant database (>800k exomes/genomes in v4) used to estimate allele frequency for rarity-based evidence.
ClinVar: NCBI database of variant–condition assertions submitted by clinical labs and curated panels.

Anchor references

Standards and guidelines for the interpretation of sequence variants (ACMG/AMP 2015) — Richards et al., Genet Med 2015
Calibration of computational tools for missense variant pathogenicity — Pejaver et al., AJHG 2022

Reading the Genome: Platforms & Modalities

Pharmacogenomics in Practice