Tumor-Normal Pairing

The core computational principle involves comparing variant allele fractions between the tumor and normal samples. A variant present in both tissues at ~50% or ~100% allele fraction is classified as a germline polymorphism. A variant observed only in the tumor sample, or at a significantly elevated allele fraction relative to the matched normal, is flagged as a somatic candidate. This subtraction eliminates the need to filter against population databases, which may miss rare private variants unique to the individual's lineage.

Unlike germline variants with fixed expected VAFs (50% heterozygous, 100% homozygous), somatic VAFs are continuous and influenced by:

• Tumor purity: The fraction of cancer cells in the sequenced sample dilutes the signal
• Clonality: Subclonal mutations present in only a subset of tumor cells exhibit low VAFs
• Copy number alterations: Amplifications or deletions distort the expected allelic ratios
• Loss of heterozygosity: Can cause somatic mutations to appear homozygous

Tumor-normal pairing enables precise estimation of cross-sample contamination. By examining homozygous single nucleotide polymorphisms in the normal sample, any reads supporting the alternate allele in the normal BAM indicate contamination from the tumor sample or another source. Tools like ContEst and Conpair use this principle to flag problematic samples before variant calling, preventing false somatic calls caused by tumor DNA leaking into the normal sample during library preparation or sequencing.

Tumor-only analysis without a matched normal relies on population frequency databases like gnomAD to filter common germline variants. This approach fails for:

• Rare private germline variants absent from population catalogs
• Population-specific polymorphisms underrepresented in reference databases
• Somatic mutations at known polymorphic sites

Paired analysis achieves >99% specificity for somatic detection, while tumor-only methods typically plateau at 85-90% specificity even with aggressive filtering.

A critical quality control step unique to paired analysis is verifying that the tumor and normal samples originate from the same individual. Concordance is assessed by comparing germline heterozygous genotypes across both samples. A mismatch rate exceeding 5% indicates a sample swap or mislabeling. Tools like NGSCheckMate and BAM-matcher compute identity-by-state metrics from the sequencing data itself, preventing catastrophic misattribution of somatic calls to the wrong patient.

A subtle failure mode occurs when circulating tumor DNA or infiltrating cancer cells contaminate the normal sample, causing true somatic mutations to appear in the normal BAM at low allele fractions. This reduces sensitivity because the subtraction algorithm incorrectly classifies these variants as germline. Advanced callers like MuTect2 and Strelka2 model this explicitly, allowing a small expected contamination rate in the normal sample to rescue borderline somatic calls that would otherwise be filtered.

The proportion of sequencing reads supporting a variant allele relative to the total read depth at that genomic locus. In tumor-normal analysis, VAF is a critical discriminator:

• Germline heterozygous variants typically show ~50% VAF in both samples
• Somatic mutations show elevated VAF in tumor but near 0% in normal
• Subclonal mutations exhibit intermediate VAFs reflecting tumor heterogeneity
• Copy number alterations distort expected VAF ratios, requiring integrated models

A systematic sequencing error where a variant allele is observed predominantly on reads from one DNA strand, indicating a technical artifact rather than a true biological mutation. Tumor-normal pairing helps identify strand bias by:

• Requiring consistent evidence across both strands in the tumor
• Comparing strand distributions between tumor and normal samples
• Flagging variants with Fisher's exact test p-values below significance thresholds
• Integrating strand bias metrics into variant quality score recalibration

Somatic copy number alterations (CNAs)—amplifications or deletions of large genomic segments—complicate variant calling by distorting expected allele frequencies. Tumor-normal pairing enables:

• Log-ratio segmentation comparing tumor to normal read depth
• B-allele frequency plots to detect loss of heterozygosity
• Integrated models that jointly call SNVs and CNAs
• Correction of VAF thresholds based on local copy number state