Panel of Normals (PoN)

The primary function of a PoN is to model and subtract recurrent technical artifacts that are not of biological origin. By identifying genomic positions where background noise consistently appears across dozens or hundreds of normal samples, the variant caller can mask these sites.

• Artifact Types Suppressed: Oxidative DNA damage (8-oxoguanine), formalin-induced deamination, and sequencer-specific cycle errors.
• Mechanism: Builds a position-specific error model using background allele frequency distributions from the normal cohort.
• Result: Prevents false positive somatic calls that would otherwise pass quality filters due to their recurrent nature.

Without a matched germline sample, a tumor-only variant caller relies on the PoN to filter out rare, population-specific germline single nucleotide polymorphisms (SNPs) that are absent from global reference databases like gnomAD.

• Ancestry Matching: The PoN must represent the genetic ancestry of the patient population to effectively tag private germline variants as 'normal'.
• Panel Size Dependency: A larger PoN (e.g., n > 50) provides greater statistical power to identify and flag low-frequency population variants.
• Filtering Logic: Variants observed in multiple normals at low allele frequencies are annotated as likely germline events and excluded from the somatic output.

The PoN defines the baseline landscape of clonal hematopoiesis of indeterminate potential (CHIP) and age-related somatic mosaicism present in the circulating white blood cells of healthy donors.

• CHIP Filtering: Common CHIP mutations (e.g., in DNMT3A, TET2, ASXL1) are captured in the PoN, preventing their misclassification as tumor-derived circulating tumor DNA.
• Signal Source Deconvolution: The PoN establishes the expected biological noise floor from hematopoietic cells, which is critical for high-sensitivity minimal residual disease detection.
• Age Correlation: The prevalence of these background mutations increases with donor age, necessitating an age-matched normal cohort for accurate subtraction.

The number of normal samples in a PoN directly impacts the sensitivity-specificity trade-off. An underpowered panel fails to suppress rare artifacts, while an excessively large panel can mask true low-frequency somatic variants.

• Statistical Basis: The PoN enables a Fisher's Exact Test or similar statistical framework to compare variant-supporting reads in the tumor case against the background error rate in normals.
• Saturation Point: Beyond a certain sample size (often 50-100), the marginal benefit of adding more normals diminishes for common artifact sites.
• Rare Artifact Detection: Detecting very rare, recurrent artifacts requires a larger panel to achieve sufficient statistical confidence.

Beyond single nucleotide variants, the PoN provides the essential read-depth baseline for detecting somatic copy number alterations (CNAs) from off-target or on-target capture reads.

• GC-Normalized Coverage: The PoN establishes the expected normalized coverage for each genomic bin, accounting for GC bias and probe efficiency.
• Log2 Ratio Calculation: Tumor coverage is divided by the PoN median coverage in each bin to calculate log2 ratios, revealing amplifications and deletions.
• Batch Correction: Principal component analysis (PCA) on the PoN coverage matrix can identify and remove latent batch effects before case sample normalization.

The computational subtraction of inherited polymorphisms from a somatic variant callset. A PoN suppresses technical artifacts, while germline filtering removes biological noise—the thousands of benign heterozygous SNPs present in every individual. This is typically achieved by comparing tumor sequence against a matched normal sample from the same patient or against population databases like gnomAD. Without this step, a somatic caller would drown in true biological variants that are irrelevant to the cancer phenotype.

The specialized algorithm that consumes the PoN to distinguish true somatic mutations from errors. Modern callers like Mutect2 and VarScan2 use the PoN as a prior probability of error at each genomic position. Key functions include:

• Local realignment around indels to suppress mapping artifacts
• Tumor-normal pair analysis to calculate log-odds of somatic vs. germline origin
• Panel-based filtering to flag positions with recurrent noise in healthy samples

A machine learning preprocessing step applied before variant calling. BQR adjusts the per-base quality scores emitted by the sequencer using empirically observed error covariates: sequencing cycle, dinucleotide context, and read group. The GATK BQSR tool builds a model of covariation and applies it to raw data, ensuring that the PoN is constructed from accurately calibrated quality scores rather than overconfident vendor defaults.

A computational strategy to exclude somatic variants originating from age-related clonal expansions in blood cells (CHIP), not the solid tumor. Since a PoN is built from healthy blood-derived DNA, it may not capture CHIP mutations common in older populations. A CHIP filter cross-references variants against known clonal hematopoiesis hotspots (e.g., DNMT3A, TET2, ASXL1) or requires a paired buffy coat sample to prevent false-positive ctDNA calls.

A normalization step that models and removes the non-linear relationship between fragment guanine-cytosine content and sequencing coverage. GC-rich regions amplify inefficiently, creating coverage troughs that mimic copy number losses in a PoN. Tools like GATK's DenoiseReadCounts apply a multiplicative correction factor derived from the PoN samples themselves, ensuring the baseline coverage profile reflects true diploid copy number, not amplification chemistry artifacts.

A random nucleotide barcode ligated to individual DNA molecules before amplification. When building a PoN from UMI-enabled libraries, the resulting consensus reads eliminate polymerase errors that occurred during library preparation. This produces a dramatically cleaner baseline—reducing the background error rate from ~1% to <0.01%—and allows the PoN to model only persistent, position-specific sequencer artifacts rather than stochastic amplification noise.