Improving protein secondary structure prediction based on short subsequences with local structure similarity

Table 2 Performnace comparison of SymPred, SymPsiPred, and PROSP on the DsspNr-25 dataset.

DsspNr-25 (8,297 proteins)	Q ₃	Q₃Ho	Q₃Eo	Q₃Co	sov	sovH	sovE	sovC
SymPred*	81.0	84.3	71.6	77.7	76.0	82.5	76.9	70.7
SymPred⁺	80.5	84.1	70.9	77.5	75.6	82.3	76.4	70.3
SymPsiPred	83.9	81.5	75.8	83.9	80.2	82.3	80.3	76.5
PROSP	75.1	79.7	67.6	71.3	68.7	77.0	73.0	63.4

Q₃Ho (Q₃Eo and Q₃Co, respectively) represents correctly predicted helix (strand and coil, respectively) residues (percentage of helix observed). sovH/E/C values are the specific SOV accuracies of the predicted helix, strand and coil, respectively. SymPred* represents the experiment result using leave-one-out cross validation and SymPred⁺ represents the experiment result using 10-fold cross validation.