Figure 8

SCOP superfamily benchmark. The figure shows how the SCOP database is divided into training and test sets. For each classifier tested on the superfamily benchmark, the sequences of the SCOP database are divided into positive and negative training and test sets. One SCOP family is used as a positive test set. The negative test set is made from one random family from each of the other superfamilies. The positive training set is made from the superfamily of the classifier, excluding the positive test set family. The negative training set is made from the other superfamilies, excluding the negative test sets.