Performance of fixed-target function predictors. Twelve function predictors were evolved with fixed targets (listed in the Methods) instead of the evolved targets used in the self-supervised approach. Each predictor is constructed from 100 independently evolved classifiers, and its "prediction score" is simply the fraction of positive classifications (out of 100). Accuracy and coverage are presented at all possible prediction score thresholds (stopping when the number of positive predictions falls below 10). In the accuracy and coverage calculations: tp = true positives, tn = true negatives, fp = false positives, fn = false negatives.