Table 8.

Comparison to other protein disorder predictors over a blind-test set. The non-redundant blind-test set contained 1,304 recent PDB chains that were unrelated to any VSL2 training sequence. The three VSL2 predictors were re-trained with all 1,327 training sequences. The prediction accuracies were calculated at default thresholds, e.g. 0.5 for VL-XT, VL3-E, VSL1, VSL2B, VSL2P and VSL2.

	SN	SP	ACC	SN S	SN L	AUC
VL-XT	56.0	77.9	67.0	55.8	53.0	74.4
VL3-E	28.4	91.4	59.9	26.8	59.7	69.2
DisEMBL	25.9	97.9	61.9	25.7	26.5	83.9
RONN	34.7	88.8	61.8	33.6	56.4	67.2
DISOPRED2	53.8	94.7	74.2	53.5	48.6	84.6
VSL1	75.6	85.8	80.7	76.1	63.6	88.5
VSL2B	73.7	80.8	77.3	74.2	63.1	84.1
VSL2P	78.4	80.4	79.4	78.5	71.6	86.7
VSL2	79.4	81.4	80.4	79.6	73.7	87.5
(a) per-chain
	SN	SP	ACC	SN S	SN L	AUC
VL-XT	54.2	79.2	66.7	54.7	52.3	74.0
VL3-E	36.5	93.9	65.2	30.5	59.2	75.9
DisEMBL	25.3	98.3	61.8	25.4	24.7	80.1
RONN	41.6	89.1	65.4	39.1	51.2	71.3
DISOPRED2	47.8	95.6	71.7	48.7	44.6	83.4
VSL1	68.3	87.4	77.9	70.9	58.3	86.2
VSL2B	67.1	82.6	74.8	68.9	60.3	82.0
VSL2P	72.7	82.5	77.6	74.2	67.1	85.3
VSL2	74.7	82.8	78.7	76.1	69.4	86.3
(b) per-residue