Abstract
Circulating tumor DNA (ctDNA) assays are commonly described by fixed variant allele fraction (VAF)-based limits of detection. However, such metrics overlook a fundamental constraint: the finite number of analyzable DNA molecules in an individual plasma sample. As a result, nominal assay sensitivity may overestimate what is physically achievable in routine clinical specimens. Genome-equivalent (GE) distributions from three independent liquid-biopsy cohorts comprising 5238 plasma samples were integrated with a Poisson-based sampling model to estimate input-limited detectability for single-locus variants. For each sample, the lower limit of detection was defined as the minimum VAF associated with a 95% probability of observing at least 1, 3, 5, or 10 mutant molecules. GE input varied widely (median 5531; interquartile range 2784-13,060). Detection at 1% VAF was theoretically achievable for nearly all samples, but performance declined sharply at lower VAFs and with increasing evidentiary thresholds. At 0.1% VAF, 72% of samples supported detection of at least one mutant molecule, compared with 46%, 35%, and 21% for thresholds of at least 3, 5, and 10 molecules. At 0.01% VAF, fewer than 10% of samples met any detection criterion. These findings indicate that many clinical plasma samples are unlikely to support reliable single-locus detection at very low VAFs, independent of assay design. Sample-aware interpretation of ctDNA results that accounts for molecular input is therefore warranted.