Figure 7

Variant calling summary for FFPE CRC samples. A) Violin plots of the variant calling thresholds (x-axis) for each sample stratified by hypothesis group: GC > AT transitions (blue right panel) versus all other substitutions (left red panel). The thresholds are higher for samples with reduced QFI estimates (y-axis) and for GC > AT hypotheses. The width in the plots correspond to the density of points in the region. B) Violin plots of metrics associated with SNP rediscovery. The transition to transversion (Ti/Tv) ratios, variants called per kilobase, and the percentage of variants annotated by dbSNP or COSMIC are shown in the panels from left to right. All metrics are independent of QFI based on univariate analysis, except the number of variants called per kilobase which is significantly negatively correlated with QFI. C) Plots of the percent variant for each of the KRAS alleles as measured by the single amplicon confirmation platform (y-axis) as a function of the percent variant found using TAS with the 1052-amplicon panel (x-axis). In this plot, the predicted and confirmed variants were called using relaxed thresholds for the 1052-amplicon panel method results with confirmation at 1% by the single-amplicon assay sequenced on the PGM (horizontal orange dashed line). D) 2x2 tables showing classification performance of KRAS variant prediction from TAS with the 1052-amplicon panel stratified by default threshold analysis (upper 2x2 table) or incorporation of confirmation testing and relaxed thresholds (lower 2x2 table). There are 41 samples considered for analysis spread across 4 KRAS alleles giving 4*41 = 164 total hypotheses tested.