Use B-factor related features for accurate classification between protein binding interfaces and crystal packing contacts

Table 3 Comparison with existing methods PISA and EPPIC.

Tested on	Methods	Prec	Sens	Spec	Acc	MCC
BNCPCS	EPPIC-core	0.98	0.76	0.99	0.90	0.79
	AvgΣB	1.00	0.85	1.00	0.94	0.88
	avgΣB*avgNo.B	1.00	0.83	1.00	0.93	0.86
Ponstingl	EPPIC-core	0.90	0.75	0.93	0.85	0.70
	AvgΣB	0.94	0.83	0.96	0.90	0.80
	avgΣB*avgNo.B	0.98	0.79	0.99	0.90	0.81
	EPPIC	0.92	0.90	0.87	0.89	0.76
	PISA	0.87	0.89	0.77	0.84	0.66
Bahadur	EPPIC-core	0.92	0.80	0.95	0.89	0.77
	AvgΣB	0.85	0.81	0.91	0.87	0.73
	avgΣB*avgNo.B	0.89	0.80	0.94	0.88	0.75
	EPPIC	0.78	0.89	0.84	0.86	0.72
	PISA	0.65	0.89	0.69	0.77	0.57

All of these methods are optimized on the DC dataset. EPPIC-core is the classifier using the number of core residues in interfaces according to the definition in EPPIC. The performance of EPPIC or PISA is borrowed from [17]. The scores of EPPIC and PISA are absent on the BNCPCS dataset.

ISSN: 1471-2105