Limit of Detection (LoD)

The absolute number of amplifiable genome equivalents (GE) extracted from plasma is the fundamental ceiling on sensitivity. LoD is mathematically constrained by Poisson sampling statistics: if a mutation is present at 0.1% VAF, a minimum of ~3,000 GE must be interrogated to reliably capture at least one mutant molecule. Low-yield extractions from limited plasma volumes directly degrade the assay's ability to resolve rare variants, regardless of downstream error correction.

The systematic noise floor of the assay, arising from polymerase errors during library amplification, oxidative DNA damage, and sequencer base-calling inaccuracies. A variant is only detectable when its signal exceeds this noise. Standard NGS workflows exhibit error rates of ~0.1–1%, which obscures low-frequency ctDNA. Reducing this rate via molecular barcoding and specialized polymerases is the primary engineering challenge for improving LoD.

Dedicated sequencing depth—the number of times a given base is read—is directly proportional to the power to resolve low-frequency variants. To confidently call a variant at 0.01% VAF, a unique depth of >10,000x is typically required after deduplication. This often necessitates targeted panel sequencing rather than whole-genome approaches, trading genomic breadth for the extreme depth necessary to achieve a clinically actionable LoD.

A stochastic phenomenon where a true variant molecule is not captured, amplified, or sequenced due to chance. This is most severe at very low input quantities and can cause false negatives even when the mean expected number of mutant molecules is >1. LoD calculations must account for the probability of dropout, often modeled via binomial or Poisson statistics, to define the concentration at which a variant is detected with 95% confidence.

Bioinformatic pipelines apply a series of filters to distinguish signal from artifact. Key steps include:

• Base Quality Recalibration (BQR): Adjusts sequencer quality scores using machine learning models of error covariates.
• Panel of Normals (PoN): Suppresses recurrent technical artifacts identified in healthy control samples.
• Germline Filtering: Subtracts inherited polymorphisms using matched normal DNA or population databases.
• Positional Bias Models: Corrects for strand-specific and cycle-specific errors. The rigor of these filters directly impacts the achievable LoD.

A random nucleotide barcode ligated to individual DNA molecules before amplification. UMIs enable computational deduplication and error correction by grouping reads sharing the same barcode, allowing the construction of a consensus sequence that cancels out random polymerase errors. Without UMIs, the practical LoD is limited by the error rate of the sequencing chemistry itself.

The percentage of sequencing reads at a specific genomic locus that carry a variant allele. LoD is often expressed as the minimum VAF at which a variant can be reliably called. Key considerations:

• A VAF of 0.1% corresponds to roughly 1 mutant molecule in 1,000 wild-type molecules
• Low VAF detection requires high library complexity and deep sequencing coverage
• VAF is influenced by both tumor shedding and total cfDNA background

A bioinformatic strategy that leverages molecular barcodes and redundant sequencing to build consensus sequences, suppressing random polymerase and sequencer errors below the variant detection threshold. By requiring that a variant be observed in multiple independent reads from the same original molecule, error rates can be reduced from ~1% to as low as 10⁻⁵.

A specialized algorithm designed to distinguish low-frequency true somatic mutations from germline variants, sequencing errors, and mapping artifacts. Modern callers use probabilistic models that integrate base quality scores, strand bias, and position-specific error rates. The performance of the variant caller directly determines whether the theoretical LoD of the assay can be achieved in practice.

A curated collection of sequencing data from healthy individuals used to model and suppress systematic technical artifacts and recurrent background noise. By characterizing the baseline noise profile of a specific assay, a PoN enables the variant caller to distinguish true low-frequency mutations from recurrent sequencing artifacts, effectively lowering the achievable LoD without increasing false positives.