Table 4.

Comparison of secondary structure prediction for three different non-homologous datasets and two different assignment types

		DSSP			SKSP+
Method		2640	1833	1975	2640	1833	1975
SPINE X	Q3	82.7±0.5	81.3±0.5	82.3±0.5	83.2±0.4	82.1±0.6	82.6±0.4
	QH	87.4	86.1	87.4	88.1	87.1	88.0
	QE	74.2	72.8	74.0	72.9	71.8	72.3
	QC	83.3	81.6	82.3	84.7	83.0	83.4
PSIPRED	Q3	80.9±0.4	80.6±0.6	81.6±0.4	80.6±0.5	80.3±0.5	80.9±0.4
	QH	79.9	79.9	80.9	79.2	79.2	79.6
	QE	73.2	73.1	74.3	71.8	71.4	72.3
	QC	86.5	85.8	86.8	86.6	87.1	87.6

Server prediction accuracy for the three different non-homologous datasets described in the text. The SPINE X server was trained on 95% of the 2640 dataset, PSIPRED was trained on its own dataset of 1999 proteins. H denotes Helix, C for Coil, and E for Sheet. Error bars were calculated by splitting all proteins randomly into 10 equally sized sets and calculating the standard deviations of the accuracies among them.