advertisement: compare things at compare-stuff.com!

Null predictions

A separate large trial was performed using library domains which were not expected to match with a non-self library fold topology. The results are shown in Figure 4.3 alongside the positive predictions. The fraction of erroneous null predictions decreases with increasing stringency of Z-score threshold. The number of mistaken null predictions with top ranking alignments with $1.4<z<1.6$ is 14 out of a total of 112 ($z<1.6$). Thus an estimate for the probability of predicting a fold with this degree of certainty, when none should be predicted, is about 13%.