Fig. 3: Resolving CH from ctDNA variants.

a Box plot showing that the variant allele frequency (VAF) distribution of ctDNA and CH variants exhibits notable overlap in context of both high and low ctDNA levels (indicated by max ctDNA VAFs above and below 10% respectively). Number of patients represented in each box plot is indicated at the top. P values are derived from the Mann-Whitney U test. b cfDNA fragment length distribution for wildtype (WT) reads and variant-carrying reads across genomic positions in the context of CH or ctDNA variants. c The left schematic provides a summary of the computational approach used to differentiate the fragment length distribution of variant versus WT fragments. The median fragment lengths for mutated and wildtype fragments were compared at each genomic position with a somatic mutation using the Mann-Whitney U test. CH mutations [middle scatter plot] and ctDNA mutations [right] with significantly different median fragment lengths compared to WT fragments are highlighted in light blue. Pie charts show the percentage of mutations with significant differences. In (b) and (c) only variants with >15 supporting reads are shown. d Random in silico downsampling of the WBC DNA sequencing data illustrating the sequencing depth needed to detect all CH variants [left] and CH variants with VAF > 1% [right] by different numbers of supporting reads. e [Left] Application of common heuristic filters (aimed at removing CH-derived variants from plasma-only sequencing data) to all detected somatic mutations, showing the number of CH and ctDNA variants that would be excluded after applying each criterion (in a theoretical scenario where matched WBC DNA was not profiled). Pie charts indicate the sensitivity and specificity of each filtering approach. [Right] Fraction of discarded and retained somatic mutations in select genes after applying the filtering criteria.