Table 2.

Performance evaluation results (ROC50). Our test set is composed of 456 families that have at least 2 related families within the same superfamily (see section 'Data sets'). For each method, we report the number of true relationships that are detected until 50 false connections occur. We repeat this analysis four times, each time with a different definition of a false positive. A break up of the relationships by relationship-type is given in Table 3 for the first type of false positive.

	Number of true	family-family	relationships	detected by
Type of false-positive	Gapped-BLAST	PSI-BLAST	prof_sim	prof_ss (predicted ss)	prof_ss (true ss)
Different superfamily, same fold ("possible" relationship)	116	146	173	178	272
Different fold, same class ("weak" relationship)	116	148	173	178	274
Different class ("suspicious" relationship)	124	153	189	209	328
Alpha ↔ Beta ("error" relationship)	214	268	308	>395	>584