In Figure 2, we show the theoretical behavior of this model using simulated data with varying base qualities. The model performs equally well for datasets where the mean certainty of the base calls is ∼80% and higher. This suggests that thresholding base calls at Phred Q20 [99% certainty (Morin et al., 2008)] or Q30 [99.9% certainty (Ley et al., 2008)] may be overly stringent.
