Fig. 1

UMI usage does not provide consistent improvement across experimental setups. A Schematic representation of signal and noise generation from the input DNA in a typical NGS experiment. B Schematic representation of the principle of noise suppression by read grouping prior to variant calling; calls recognized as noise (i.e. not found consistently across reads derived from both strands within a read group) are ignored during read collapsing. C Schematic depiction of a double-stranded UMI system (top) and its use to identify 2 input molecules from 8 sequencing reads (bottom). D Schematic representation of the experimental setup used to investigate variant calling performance in different types of input material; 3 types of input (FF, FFPE and cfDNA) were generated from mixtures of the reference samples GM12891 and GM12892 at various ratios (see the “Methods” section); the mixes were used to generate libraries that were subsequently enriched using hybridization capture with a panel targeting 110 variants and then sequenced; after sequencing and read alignment, variant calling was performed using read mapping positions alone or using mapping positions and UMI information. E Sensitivity of duplex consensus-based variant calling (using a threshold of at least 2 duplexes to call) for 110 known variants present at VAF of 0.2% in three types of input material (12.5 ng cfDNA, 100 ng FF DNA, 25 ng FFPE DNA) with read grouping either using mapping positions alone or using mapping positions plus UMI information