While the SNVMix2 model eliminates the need for thresholding through probabilistic weighting, we explored the effect of applying thresholds to the SNVMix2 input in order to identify a practical balance between pure weighting and thresholding. We compared the results of thresholding mapping qualities at (Q0, Q5, Q10, Q20, Q30, Q40 and Q50) and concomitantly thresholding base qualities at (Q0, Q5, Q10, Q15, Q20 and Q25) and running SNVMix2 on the resulting data from both 10× and 40× genomes. As shown in Table 2, thresholding the mapping qualities at Q50 and base qualities at Q20 (mQ50_bQ20) at 10× performed better than all other 10× runs (F−measure 0.8441). For the 40× data, thresholding the mapping qualities at Q50 and base qualities at Q15 (mQ50_bQ15) performed best over all runs (F-measure 0.8658). (See Supplementary Table S1 for all results from runs in increments of Q1 base quality thresholds.) This suggests that previously reported base quality thresholds may be too stringent. Furthermore, when used with stringent mapping quality thresholds, the SNVMix2 model can effectively use the base qualities by probabilistic weighting to confer higher
