Fig. 4
Effects of UMI usage on signal and noise. A Plot displaying VAF for detected true-positive (red) or false-positive (blue) variants, for variant calling without read grouping (“no grouping”) or using mapping position and UMI consensus ("with UMIs”); data is displayed for 3 types of input material (expected VAF = 0.5%). B ROC curve for variant calling performance in cfDNA with expected VAF = 0.2%, without read grouping or using mapping position and UMI consensus. C Plots displaying area under the ROC curve (AUC), using mapping positions alone (cyan) or mapping position and UMI information (red), for 3 types of input material and a range of expected VAF values. D Schematic representing the experimental approach for investigation of signal loss (see the “Methods” section). FASTQ files from samples with targeted variants at a given VAF (“Original sample”) were merged with FASTQ files of background DNA lacking the variants (“Negative sample”). The number of variant-supporting molecules (NALT) was calculated for each dilution. An example VAF and input amounts are shown in the figure. E Fraction of ALT molecules lost (y axis) due to collisions as a function of input amount (x axis), for samples with different initial VAF values. Plots on top show the count of ALT molecules for each targeted variant in undiluted samples