Table 3. Comparison of the agreement between different classifications.
Superfam. | Folds | |
SCOP vs. CATH | 0.84 | 0.48 |
Automatic (AL) vs. SCOP | 0.54 | 0.69 |
Automatic (AL) vs. CATH | 0.58 | 0.32 |
AL (max) vs. SCOP | 0.65 | 0.79 |
AL (max) vs. CATH | 0.64 | 0.63 |
Automatic (SL) vs. SCOP | 0.24 | 0.48 |
Automatic (SL) vs. CATH | 0.28 | 0.70 |
SL (max) vs. SCOP | 0.51 | 0.67 |
SL (max) vs. CATH | 0.51 | 0.80 |
The agreement is evaluated through the weighted kappa parameter, Eq. (19). The first line compares superfamilies and folds from SCOP and CATH. In the two following lines, the automatic classification at the stop point obtained with average linkage (AL) is compared with SCOP and CATH, respectively, at the levels of superfamilies and folds. The two following lines compare the expert classifications with the AL classification at the points where their weighted kappa is maximum. The four last line are the same, but using as clustering algorithm single linkage (SL), which gives a much stronger agreement with CATH than with SCOP at the fold level, consistent with the fact that CATH uses single linkage.