Table 6. Summary of results on the test datasets TEST30_ION, TEST30_VLG, and TEST60_VGS.

Dataset	Prediction outcome	Method	Fmeasure	MCC	Accuracy	Q4
TEST30ION	Ion-channel vs. non-ion channel	VGIchan	63.0	0.49	72.7	NA
		Lin et al. 2011	81.7	0.63	80.8	NA
		BLAST	64.3	0.56	74.7	NA
		PSIONplus	85.4	0.71	85.4	NA
		Confidence interval of PSIONplus	86.0(±3.7)	0.73(±0.07)	86.3(±3.3)	NA
TEST30VLG	Voltage-gated vs. ligand-gated	Lin et al. 2011	76.6	-0.06	63.3	NA
		BLAST	77.6	0.23	68.3	NA
		PSIONplus	77.6	0.23	68.3	NA
		Confidence interval of PSIONplus	78.1(±6.1)	0.22(±0.15)	68.7(±7.4)	NA
TEST60VGS	Potassium	Lin et al. 2011	87.6	0.74	86.6	NA
		BLAST	91.6	0.83	90.8	NA
		PSIONplus	94.8	0.90	94.6	NA
	Anion	Lin et al. 2011	86.7	0.85	95.4	NA
		BLAST	86.7	0.85	95.4	NA
		PSIONplus	88.1	0.87	95.8	NA
	Calcium	Lin et al. 2011	73.7	0.67	89.5	NA
		BLAST	91.1	0.90	96.7	NA
		PSIONplus	92.0	0.90	96.7	NA
	Sodium	Lin et al. 2011	90.5	0.90	98.3	NA
		BLAST	93.0	0.93	98.7	NA
		PSIONplus	93.0	0.93	98.7	NA
	Average over	Lin et al. 2011	84.6	0.79	92.4	84.9
	all subtypes	BLAST	90.6	0.88	95.4	90.8
		PSIONplus	92.0	0.90	96.4	92.9
		Confidence interval of PSIONplus	91.9(±2.1)	0.90(±0.03)	96.4(±0.9)	92.9(±1.7)

Results of PSIONplus are compared with VGIchan on the TEST30_VLG dataset, and with the method by Lin et al. and BLAST on all datasets. Best MCC, F_measure and accuracy values for each dataset are shown in bold. Confidence intervals are obtained by computing average and standard deviations (shown in brackets) of 10 repetition of the test where in each repetition we randomly select 50% of test data set. NA means “not applicable”; for the two-class classification the Q₄ equals accuracy.