Penetrance

Penetrance is the proportion of people carrying a particular genotype who show the associated phenotype. It is a property of a variant measured in a population, not a property of the variant in isolation, and the population you measure it in changes the answer by an order of magnitude.

100% penetrance (complete penetrance) means every carrier observed developed the trait. Huntington disease with a large CAG expansion is the standard example. High penetrance usually means something above roughly 50% lifetime risk. Low or reduced penetrance means most carriers never develop the phenotype. Incomplete penetrance is the general term for anything under 100%.

How it works

The formula is trivial. The denominator is not.

penetrance = P(phenotype | genotype)
           = affected carriers / all carriers

Ascertainment is the whole problem. Classical penetrance estimates come from families referred to a genetics clinic because someone was already sick. That conditions the denominator on having an affected relative, which inflates the estimate. Population-based estimates invert the design: genotype an unselected biobank, then look up who has the phenotype in linked health records.

The gap is large. Analyses of deleterious clinical variants in unselected health-system cohorts find penetrance for many actionable variants well below the textbook family-study figures.¹ In gnomAD, a systematic review of clinically relevant variants across more than 800,000 individuals shows a substantial set of variants classified as pathogenic appearing at frequencies incompatible with the stated disease prevalence, which forces the penetrance estimate down.² For BRCA1 and BRCA2 specifically, a defined subset of pathogenic variants behaves as reduced-penetrance alleles and is now reported differently in clinical testing, because assigning them the standard high-penetrance risk figure overstates the risk.³

Mechanistically, incomplete penetrance has several sources. Modifier variants in cis or trans: the canonical case is CFTR, where the disease consequence of the T5 allele depends on the length of the adjacent polymorphic (TG)m tract on the same chromosome.⁴ Stochastic gene expression: in C. elegans, variation in expression of a single gene across genetically identical animals predicts which individuals show the mutant phenotype.⁵ Environment and common polygenic background: in CADASIL, vascular risk factors and polygenic burden shift the penetrance of NOTCH3 variants.⁶ And in the recurrent CNV space, the definition itself has been reworked, because naive carrier counts conflate de novo and inherited events and need an explicit correction.⁷⁸

In your own data

Start from the VCF, not from a consumer report.

1. Annotate. Run your GRCh38 VCF through VEP or bcftools csq, and join against the ClinVar VCF (clinvar.vcf.gz, versioned) on CHROM/POS/REF/ALT. Do not join on rsID alone: multi-allelic sites and left-alignment differences will silently mismatch. Normalize first: bcftools norm -f GRCh38.fa -m -any -c s.
2. Filter on CLNSIG=Pathogenic or Pathogenic/Likely_pathogenic, then check CLNREVSTAT. One-star submissions with no assertion criteria are noise. Require at least two stars before spending time on a variant.
3. Look up the allele frequency in gnomAD v4, per-ancestry not global. A variant present at 1 in 400 in your ancestry group cannot cause a disease with prevalence 1 in 50,000 at full penetrance. That arithmetic is the fastest sanity check you can run, and it is the core of the gnomAD penetrance work.²
4. Check your genotype quality at the site. GQ under 20, DP under 10, or an allele balance far from 0.5 on a het call means the call may be wrong before penetrance is even relevant. Pull the region in IGV from the CRAM.
5. Separate penetrance from expressivity. Expressivity is how severe the phenotype is among those affected. Two different axes, often confused, and a variant can be low-penetrance and highly variable in expression at once.⁹

Common mistakes: quoting a family-study penetrance figure for a variant found incidentally in your own genome; treating “Pathogenic” as “will happen”; ignoring that most published penetrance figures are lifetime risk by age 80, not risk this year; assuming a short-read caller found your repeat expansions or large CNVs at all (it usually did not, use ExpansionHunter and a dedicated CNV caller).

Anything you find here needs a clinical geneticist to confirm and interpret. Clinical-grade confirmation of a research or direct-to-consumer call is a separate assay, and the decision about what to do next belongs with a clinician.

Limitations

Penetrance estimates from biobanks inherit their own biases: healthy-volunteer effects, truncated follow-up (a 55-year-old carrier has not finished being at risk), and ICD-code phenotypes that miss mild disease. Ancestry coverage is uneven, so a penetrance figure derived largely from European-ancestry cohorts may not transfer. And for most individual variants the carrier count in any cohort is single digits, so the confidence interval spans nearly the full 0 to 1 range. Treat any point estimate without an interval as a placeholder.

Woolf Software builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.