Predicting protein-ATP binding sites from primary sequence through fusing bi-profile sampling of multi-view features

Table 4 Combination of different feature groups and performance comparison based on leave-one-out jackknife tests on ATP168 and ATP227

	Composition of different features	Sen (%)	Spe (%)	Acc (%)	AUC
ATP168	PSSM	76.47	76.75	76.74	0.8466
	LogisticPSSM^a	77.22	77.55	77.54	0.8553
	LogisticPSSM + Bipro-aa	77.45	77.71	77.70	0.8609
	LogisticPSSM + Bipro-dis	77.45	77.76	77.74	0.8601
	LogisticPSSM + Bipro-sa	77.45	77.96	77.93	0.8619
	LogisticPSSM + Bipro-ss	77.84	78.20	78.18	0.8638
	LogisticPSSM + Bipro-all^b	77.75	78.16	78.13	0.8633
ATP227	PSSM	79.23	79.63	79.61	0.8678
	LogisticPSSM^a	79.98	80.14	80.14	0.8788
	LogisticPSSM + Bipro-aa	80.30	80.54	80.53	0.8822
	LogisticPSSM + Bipro-dis	80.36	80.53	80.53	0.8818
	LogisticPSSM + Bipro-sa	80.54	80.76	80.75	0.8843
	LogisticPSSM + Bipro-ss	80.57	80.85	80.83	0.8851
	LogisticPSSM + Bipro-all^b	80.54	80.60	80.70	0.8816

^a LogisticPSSM means PSSM profile normalized by logistic function.
^b all means the combination of four feature groups: Bipro-aa, Bipro-dis, Bipro-sa, and Bipro-ss.

ISSN: 1471-2105