Can you tell germline from somatic just by looking at VAF?

Not reliably. A heterozygous germline variant sits near 50% VAF, but subclonal somatic variants can sit as low as 2 to 5%, and clonally expanded somatic variants from CHIP or a high-purity tumor can sit at 30 to 50%, overlapping directly with the germline range.

Why do you need a matched normal sample to call somatic mutations?

The matched normal is the only direct evidence of what's inherited in this specific patient. Variants detected in the tumor but absent from the patient's own normal tissue are the candidate somatic set; population databases only tell you what's common across other people, not what this patient inherited.

What is CHIP and why does it confuse germline vs somatic calls?

CHIP (clonal hematopoiesis of indeterminate potential) is somatic mutation and clonal expansion in blood stem cells that increases with age, reaching 10% or more prevalence in people over 70. Because CHIP variants in genes like DNMT3A and TP53 can reach 10 to 40%+ VAF in blood, they can be misread as germline cancer-predisposition variants or misattributed to a solid tumor in liquid biopsy panels.

Does tumor-only sequencing reliably separate somatic from germline?

Not as reliably as a matched-normal design. Tumor-only pipelines filter candidate somatic variants against population allele-frequency databases, but those databases are incomplete, so rare real germline variants get miscalled as somatic when there's no patient-matched normal to check against directly.

Glossary · Genomics and Variants

Germline vs somatic variants

The label you assign a variant decides whether it goes in a genetic counseling report or a tumor board slide, and VAF alone won't tell you which.

By Ming "Tommy" Tang, Director of Bioinformatics in Big Pharma · Reviewed September 2026 · 3 min read

Also: somatic mutation, germline mutation

Definition

A germline variant is inherited through the germ line, present in essentially every cell of the body from conception onward, and transmissible to offspring; a heterozygous germline SNV shows up at roughly 50% allele frequency in a diploid genome, homozygous near 100%. A somatic variant arises after conception in a subset of cells during an individual's lifetime, is not present in the germ line, and is not passed to children. Because a somatic variant only exists in the cells descended from the one where it first occurred, its allele frequency in a bulk sample depends on what fraction of the sampled cells carry it, driven by tumor purity, clonal expansion, or subclonal architecture rather than Mendelian inheritance.

You hit this distinction the moment a variant caller hands you a VCF and someone asks "is this inherited or did the tumor pick it up?" The answer decides two very different next steps: a germline finding routes to genetic counseling and family risk assessment, a somatic finding routes to a tumor board discussion about targeted therapy. Get the label wrong and you send a family for hereditary cancer testing over a variant their tumor made on its own, or you miss a heritable risk because you assumed everything in a cancer panel is tumor-derived.

The distinction sounds obvious in a textbook: germline is inherited and in every cell, somatic is acquired and only in some. In real sequencing data it's messier, because allele frequency, sample source, and caller design all conspire to blur the line, especially in blood-derived "normal" samples that aren't as clean as their name implies.

Why it matters

The textbook heuristic, germline sits near 50%/100% VAF, somatic sits lower and messier, breaks down constantly in practice, and the break has clinical teeth. Clonal hematopoiesis of indeterminate potential (CHIP) puts somatic mutations in genes like DNMT3A, TET2, ASXL1, and TP53 into blood stem cells that expand with age, reaching VAFs of 10 to 40%+ in people over 70. A DNMT3A variant at 30% VAF in a 70-year-old's "normal" blood sample looks nothing like background noise; read naively it can be misclassified as a germline cancer-predisposition variant, or, if that blood is the source of a liquid biopsy, wrongly attributed to the solid tumor and used to justify a targeted therapy it has nothing to do with.

The inverse failure shows up in tumor-only pipelines that skip a matched normal and instead filter against population databases like gnomAD. Because those databases are incomplete, rare but real germline variants that aren't yet catalogued get called as somatic. This is exactly why tumor-normal paired sequencing remains the reference approach: variants detected in the tumor but absent from the patient-matched normal are the candidate somatic set, and everything else is presumptively germline.

Where people get it wrong

The mistake practitioners make is trusting VAF as a clean discriminator instead of a probabilistic signal that depends on tumor purity and clonal fraction. A subclonal somatic variant can sit at 2 to 5% VAF, well below a naive threshold, while a CHIP-derived somatic clone or a high-purity tumor's clonal driver can sit at 30 to 50% VAF, squarely in "looks germline" territory. The second version of this mistake is architectural: treating tumor-only calling (filter against a population database) as a substitute for tumor-normal calling. A population database tells you a variant is common in the general population, not that it's absent from this patient's germline; anything rare or private gets misrouted to the somatic bucket by default. Matched-normal sequencing solves this directly because it asks the only question that actually settles it, is this variant in the patient's own non-tumor tissue, yes or no.

A concrete example

Run Mutect2 on a tumor-normal pair, then check whether a candidate variant is somatic by confirming it's absent from the normal BAM, not by eyeballing its VAF. The germline-resource and panel-of-normals flags encode population and recurrent-artifact information, but the matched normal is what actually earns the somatic call.

bash

gatk Mutect2 \
  -R reference.fasta \
  -I tumor.bam \
  -I normal.bam \
  --germline-resource gnomAD.vcf \
  --panel-of-normals pon.vcf \
  -O somatic.vcf.gz

# then filter on quality before trusting any call
vcffilter -f 'QD > 4 & FS < 30 & MQ > 50' somatic.vcf.gz

Related terms

Questions people ask

What's the actual difference between a germline and a somatic variant?: A germline variant is inherited, present in the germ line and every somatic cell, and can be passed to children. A somatic variant arises after conception in a subset of cells, is confined to the tissue it arose in (or its clonal descendants), and cannot be inherited by offspring.
Can you tell germline from somatic just by looking at VAF?: Not reliably. A heterozygous germline variant sits near 50% VAF, but subclonal somatic variants can sit as low as 2 to 5%, and clonally expanded somatic variants from CHIP or a high-purity tumor can sit at 30 to 50%, overlapping directly with the germline range.
Why do you need a matched normal sample to call somatic mutations?: The matched normal is the only direct evidence of what's inherited in this specific patient. Variants detected in the tumor but absent from the patient's own normal tissue are the candidate somatic set; population databases only tell you what's common across other people, not what this patient inherited.
What is CHIP and why does it confuse germline vs somatic calls?: CHIP (clonal hematopoiesis of indeterminate potential) is somatic mutation and clonal expansion in blood stem cells that increases with age, reaching 10% or more prevalence in people over 70. Because CHIP variants in genes like DNMT3A and TP53 can reach 10 to 40%+ VAF in blood, they can be misread as germline cancer-predisposition variants or misattributed to a solid tumor in liquid biopsy panels.
Does tumor-only sequencing reliably separate somatic from germline?: Not as reliably as a matched-normal design. Tumor-only pipelines filter candidate somatic variants against population allele-frequency databases, but those databases are incomplete, so rare real germline variants get miscalled as somatic when there's no patient-matched normal to check against directly.

Sources

Precise identification of somatic and germline variants in the absence of matched normal samples — OncoTOP decision-tree accuracy and VAF limits of detection
A computational approach to distinguish somatic vs. germline origin of genomic alterations — SGZ method modeling tumor purity and copy number for classification
GATK Mutect2 Documentation — Mutect2 tumor-normal command and prior probability settings
Genomic variants concurrently listed in somatic and germline mutation databases — Genes like TP53 causing both hereditary and CHIP-driven mutations