Figure 3

Average ratio of true positives to the sum of true and false positives versus E -values.

In this plot, we show the assessment of the accuracy of calibrated E-values. The ratio TP/(TP + FP) is plotted against E-value, see text for more details. The black solid curve corresponds to the theoretical line given by Eq. (1). In the left panel, we plot the results from four E-value-reporting search methods using the original E-values prior to calibrations. Not all the experimental curves are close to the theoretical curve. In the right panel, the calibrated E-values are used and one sees that the four curves from the left panel now form a much tighter bundle. Furthermore, even for methods that do not report E-value directly, except perhaps for ProbID we see they also fall within the bundle. The error bars in these plots are suppressed for clarity. They are about the size of the symbol at the low E-value end, and become larger (four to six times the symbol sizes) near large E-value end.