Figure 6.

Iteration improves sensitivity of DDiMAP variant identification and coverage of simulated Illumina HiSeq Data. (A) Precision-Recall curves were generated by varying DDiMAP primary filter thresholds from 100 to 800 ppm and co-varying variant identification thresholds at 4× primary filter levels (see Additional file 1 for details). Performance is shown for initial and final iterations. A peak performance F1-score for each case is shown in matching color. Significant improvement in sensitivity is obtained with iteration for SHRiMP2 and CUSHAW3, with improvement in precision for CUSHAW3 as well. Novoalign, using default settings, had slightly improved sensitivity. DDiMAP attained its peak performance for all three mappers at a threshold of 300 ppm. (B) Coverage for SHRiMP2 (S) and CUSHAW3 (C) at the initial iteration, after the third iteration, and at the final iteration are shown. In these cases the final enhanced reference sequences contained the private founder clone sequence without creating false positive variants at any frequency. Coverage using Novoalign (N) did not improve with iteration.