HemeBIND: a novel method for heme binding residue prediction by combining structural and sequence information

BMC Bioinformatics

Table 5 Performance of different prediction models on independent test set

Subset	Model^a	Recall (%)	Precision (%)	Accuracy (%)	F1-score (%)	MCC
Single-heme	STR	76.05	28.72	77.06	41.69	0.366
	STR_RFP	72.10	31.44	80.03	43.78	0.382
	SEQ	58.86	39.35	85.78	47.17	0.404
	STR+SEQ	52.28	52.75	89.80	52.51	0.468
	STR_RFP+SEQ	50.54	55.78	90.34	53.03	0.477
Multi-heme	STR	87.07	51.48	69.66	64.70	0.454
	STR_RFP	86.56	52.60	70.80	65.43	0.466
	SEQ	63.14	59.85	74.70	61.45	0.427
	STR+SEQ	58.55	69.53	78.57	63.57	0.489
	STR_RFP+SEQ	58.25	70.18	78.76	63.66	0.493
All	STR	80.13	34.93	75.83	48.65	0.412
	STR_RFP	77.45	37.72	78.50	50.73	0.431
	SEQ	60.44	45.36	83.94	51.83	0.431
	STR+SEQ	54.60	58.34	87.94	56.41	0.495
	STR_RFP+SEQ	53.39	60.82	88.42	56.87	0.504

^aSTR, RFP and SEQ denote structure-based classifier, reducing false positives and sequence-based classifier respectively.

ISSN: 1471-2105