Table 3. Each individual score's separation parameter and performance as a one-score classifier.

Score	Type	Separationa	Optimal thresholdb	Accuracy (%)
				Remaining	Comparable	Close	SF	FD	CL	RT
Dali	Intra	0.42	0.75	67.69	70.00	73.87	24.60	93.18	90.04	92.15
Daliz	Intra	0.36	0.52	70.77	63.08	71.17	20.87	95.57	97.35	98.55
GDT_S	Inter	0.28	0.66	66.92	63.08	67.57	12.95	96.35	98.99	99.36
TM score	Inter	0.42	0.56	74.62	63.08	83.78	69.96	54.92	96.67	99.59
RMSD	Inter	0.38	0.77	70.00	60.00	86.49	59.02	77.08	96.31	98.32
AHM	Inter	0.39	0.64	73.85	63.85	77.48	74.44	46.05	80.45	85.74
LBa	Intra	0.29	0.36	61.54	56.92	96.40	84.03	37.95	47.02	60.04
LBb	Intra	0.24	0.48	58.46	60.00	92.79	87.04	29.22	48.05	64.52
LHM	Inter	0.11	0.52	63.08	52.31	74.77	21.07	94.03	99.91	99.98
Id	Pair	0.41	0.06	80.00	51.54	88.29	46.40	81.42	86.13	87.44
Blosum	Pair	0.53	0.07	79.23	56.92	95.50	76.21	52.76	72.37	77.09
Compass-like	Profile	0.86	0.18	86.15	76.92	93.69	45.06	82.40	95.72	97.87
Correl	Profile	0.70	0.19	83.08	70.77	91.89	57.69	75.96	92.26	95.33
HHsearch	Profile	0.75	0.02	86.15	66.15	90.09	70.17	58.21	87.10	95.08
SVM scorec		0.93	0.40	94.62	76.15	90.99	75.04	56.59	80.63	84.01

^aSeparation of a score is calculated by the following equation: separation=(μ_h − μ_a)/(σ_h + σ_a), where μ_h, σ_h, μ_a, and σ_a are the mean and standard deviation for homologs and analogs in the “remote” data set, respectively.

^bThe optimal threshold for a single-score classifier is found by scanning a wide range of thresholds and identifying the one at which the accuracy on “remaining” is the highest.

^cThis is the SVM prediction score given by the final classifier.