Classification of proteins in SCOP and CATH using predicted helix interactions.

Helix interactions were predicted using the threshold combination C = 9 (network NN4) and C = 15 (network NN4-D), see Materials and Methods. HISS scores were calculated with and without weighting edges.
aHISS scores were calculated both without weighting helix interactions (uw) and with up weighting interactions involving >15 residue contacts by a factor 1.5 (w).
bAvg(HISS)same: average HISS score for proteins classified to the same fold in SCOP and CATH.
cAvg(HISS)diff: average HISS score for proteins classified to different folds in SCOP and CATH.
dAUC: area under the curve describing how well proteins with the same fold can be differentiated from proteins with different folds (AUC = 0.5 would correspond to a random prediction).
eScore: HISS score threshold used to identify proteins with the same helix architecture. For both weighted and unweighted HISS scores, two thresholds were chosen such that the specificity of the obtained classifications most closely approached either 80% or 90%.
fSensitivity: Fraction of all protein pairs with the same SCOP/CATH fold annotation having a HISS score above the specified threshold.
gSpecificity: Fraction of all protein pairs with different SCOP/CATH fold annotation having a HISS score below the specified threshold.