| advertisement: compare things at compare-stuff.com! |
It is essential that sets of folds used in class prediction experiments do not contain homologous pairs. If a query sequence has a significant match to a sequence in the dataset (using standard sequence alignment tools) then clearly its secondary structural class (and architecture and topology) can be assigned unambiguously to that of the homologous sequence of known structure. The CATH dataset used here contains only homologous superfamily representatives. None of these should share significant pairwise sequence similarity, however a few pairs are homologous and are corrected in subsequent releases of CATH.