Figure 3.

Quality calibration curves for AYB and Ibis. Quality calibration curves for AYB (left) and Ibis (right) on the Ibis Test data set of 200K 50 cycle reads. For each base-caller, two graphs are superimposed: a line chart with error bars showing how the empirical quality changes with estimated quality (left axis) and a histogram of how many bases have a given estimated quality (right axis, as a proportion of mapped bases). The error bars represent 99% confidence intervals, obtained by transforming the Wilson interval [32] for the proportion into a quality score. On each graph, straight lines are shown representing perfect correlation between estimated and empirical qualities (solid) and the best linear weighted fit (dashed).