Figure 4

Independent sequencing of reference DNA mixtures exposes true variants and platform-specific false positives. A) The percent variant of the 6 reference DNA mixtures as measured from sequencing using the 1052-amplicon (x-axis) and the AmpliSeq (y-axis) panels. The expected variants are green circles and all others are identified with a red X. Variants detected with both workflows fall close to the y = x line, while platform specific false positives (FP) are plotted along the x = 0 axis (AmpliSeq false positives) and y = 0 axis (1052-amplicon panel false positives). B) A heatmap of the f-measures (the harmonic mean of sensitivity and PPV) from the two panels for variant calling using different thresholds for percent variant. Each point in the heatmap is colored by the f-measure with green positions having the highest performance. Positions with an f-measure <0.5 are colored gray to prevent color saturation for higher values. The variant calling performance of the 1052-amplicon panel alone is maximized at 6% (label C), but performance can be significantly improved by relaxing threshold constraints for variant calling and considering information from both panels or workflows (label D). In this figure, the systematic variants are not removed from either platform so that the analysis can further illustrate the complementary behavior of each platform. The AmpliSeq panel alone would have performance maximized at position 27.5% (label E). C) The 2x2 table showing maximum variant calling performance for samples sequenced after target enrichment with the 1052-amplicon panel. D) The 2x2 table showing maximum performance for samples sequenced after target enrichment with both the AmpliSeq and 1052-amplicon panels. E) The 2x2 table showing maximum performance for samples sequenced after target enrichment with the AmpliSeq panel only.