Figure 2

Threshold and cross-validation filters effectively remove signal noise while retaining true variant calls. (A) Negative control plasmid fragments (pBluescsript II KS) were spiked into FL specimen pooled amplicons at 1/10 the concentration of the individual targeted gene amplicons. Aggregating the aligned read data from the 12 FL specimens after mapping with BFAST, resulted in an average location coverage of 45,000×, in which there were 989 observed raw candidate SNV calls discovered (black line). Application of either the bidirectional minimum word threshold frequency filter at 750 ppm (red diamonds) or by cross-verification (green squares) alone dropped the candidate SNV call counts to less than 10% of the raw calls, while application of both filters had a synergistic effect, eliminating >99.5% of the initial calls (4/989 - blue triangles). (B) The IGH data from the 10 FL specimens was aggregated in a similar manner, resulting in 45,426 raw candidate SNV calls (black line) with a lesser reduction due to cross-verification (18,684 remaining or 41% - green squares), a similar reduction due to thresholding (4,714 or 10% - red diamonds) and a combined reduction to 1,948 final SNV calls (>4% - blue triangles) following application of both threshold and cross-verification filters. Note that the combination of filters retains the vast majority of SNV calls which were present at frequencies >1%. Reads were mapped to the Sanger-level sequence of the clonal IGH from each FL specimen and represents SHM generated variation around the clonal sequence.