Germline Filtering

The gold-standard approach where a patient's tumor sample is compared directly against their own non-tumor tissue, typically peripheral blood mononuclear cells (PBMCs) or buccal swabs.

• Variants present in both samples are classified as germline polymorphisms and removed
• Variants unique to the tumor are retained as somatic candidates
• Eliminates the need for population frequency databases
• Critical for detecting rare familial variants that may be absent from public databases

When a matched normal sample is unavailable, variants are filtered against large population catalogs such as gnomAD, 1000 Genomes, and ExAC.

• Any variant with a population allele frequency above a defined threshold (typically >1%) is flagged as likely germline
• Requires careful consideration of ancestry-matched sub-population frequencies
• Risk of misclassifying founder mutations common in specific ethnic groups
• Cannot distinguish rare private germline variants from true somatic events

Germline variants in a pure tumor sample are expected to appear at allele frequencies consistent with heterozygous (50%) or homozygous (100%) states, adjusted for tumor purity and copy number.

• Somatic variants often present at lower, subclonal frequencies
• Tumor purity estimation is essential for accurate expected frequency calculation
• Loss of heterozygosity events can cause germline variants to appear at unexpectedly high frequencies
• Statistical models incorporate local copy number and stromal admixture

A curated collection of sequencing data from a cohort of healthy individuals, processed through the identical laboratory and bioinformatic pipeline as the tumor samples.

• Models and suppresses systematic sequencing artifacts and recurrent technical noise
• Captures platform-specific errors that population databases miss
• Requires periodic updating as reagents and instruments change
• Essential for tumor-only variant calling workflows where matched normals are unavailable

Age-related clonal expansions of hematopoietic stem cells introduce somatic mutations into blood that are not tumor-derived but appear in plasma cfDNA.

• Variants in genes such as DNMT3A, TET2, and ASXL1 are classic CHIP mutations
• Matched white blood cell sequencing can distinguish CHIP from tumor-derived ctDNA
• Failure to filter CHIP variants leads to false-positive tumor mutation calls
• Particularly problematic in older patient populations where CHIP prevalence exceeds 10%

True germline variants are expected to appear on both forward and reverse sequencing strands with balanced allele representation, while artifacts often exhibit strand orientation bias.

• Fisher's exact test or beta-binomial models quantify strand imbalance
• Oxidative damage from sample preparation creates 8-oxoguanine artifacts with characteristic G>T transversions
• FFPE-induced deamination produces C>T artifacts that mimic true mutations
• Computational filters apply strand bias thresholds to reject spurious calls

The gold-standard method where a somatic variant caller compares tumor tissue or plasma against a matched normal sample (typically blood or buccal swab) from the same patient. Variants present in both samples are classified as germline and removed. This approach directly accounts for individual genetic ancestry, eliminating the need to rely on population databases. The subtraction relies on precise Variant Allele Frequency (VAF) ratios and statistical models that distinguish heterozygous germline SNPs (expected ~50% or 100% VAF in the normal) from subclonal somatic mutations.

A computational shortcut used when a matched normal sample is unavailable. Variants are intersected against large-scale population catalogs like gnomAD, 1000 Genomes, or ExAC. Any variant with a population allele frequency above a defined threshold (e.g., >1%) is flagged as a likely benign polymorphism and removed from the somatic callset. This method is prone to false positives in under-represented ethnic groups and can erroneously remove founder mutations. A Panel of Normals (PoN) serves as a technical complement to filter recurrent sequencing artifacts.

A critical biological confounder in liquid biopsy. Age-related clonal expansions in hematopoietic stem cells produce somatic mutations in blood cells that are shed into the plasma as ctDNA. Without a dedicated Clonal Hematopoiesis Filter, these variants are misattributed to a solid tumor. Germline filtering alone cannot resolve CHIP, as these are true somatic events. Mitigation requires sequencing the matched blood buffy coat to its own depth or using machine learning classifiers trained on CHIP-associated gene lists (e.g., DNMT3A, TET2, ASXL1).

A rapid filtering strategy leveraging the expected Variant Allele Frequency distribution of germline events. True heterozygous germline SNPs cluster tightly around 50% VAF in a diploid normal sample, while somatic mutations in ctDNA often exhibit low, variable VAFs (e.g., 0.1%–10%). Algorithms apply Bayesian mixture models to separate these distributions. This method is computationally efficient but fails for copy-number altered regions where germline VAF deviates from 50%, and for high-VAF somatic mutations in samples with high tumor content.

An orthogonal validation technique that examines the trinucleotide context of each variant. Germline polymorphisms exhibit a distinct mutational spectrum shaped by population genetics and DNA repair, while somatic mutations carry signatures of specific Mutational Signature processes (e.g., APOBEC, smoking, UV). By computationally deconvolving the aggregate mutation spectrum, variants inconsistent with the expected somatic signatures can be flagged for review. This method provides a biological rationale for filtering that is independent of allele frequency or database membership.

A quality control step that removes technical artifacts mimicking germline variants. True variants are expected on both forward and reverse sequencing reads. Strand bias occurs when a variant is disproportionately supported by reads from a single orientation, indicating an artifact from oxidative damage (8-oxoguanine) during library preparation or sequencing. Germline filtering pipelines integrate Base Quality Recalibration (BQR) and strand-bias metrics (e.g., Fisher's Exact Test) to purge these systematic errors before biological classification begins.