Machine learning approach to predict protein phosphorylation sites by incorporating evolutionary information

BMC Bioinformatics

Table 7 Prediction performance of the system when testing with the independent benchmark dataset.

Residue	W	Ac(%)	Sn(%)	Sp(%)	Mcc
	7	61.34	75.83	61.16	0.08
	9	67.82	67.77	67.82	0.08
S	11	68.44	70.14	68.42	0.09
	13	67.77	67.77	67.77	0.08
	15	64.96	72.04	64.88	0.08
	7	68.34	64.71	68.37	0.06
	9	73.81	58.82	73.93	0.07
T	11	71.73	61.18	71.81	0.06
	13	69.60	64.71	69.64	0.07
	15	69.87	67.06	69.90	0.07
	7	64.95	76.29	64.75	0.11
	9	62.40	79.38	62.09	0.11
Y	11	61.48	77.32	61.19	0.10
	13	61.88	80.41	61.54	0.11
	15	64.83	76.29	64.62	0.11

From this table it is evident that using ratio 1:1 shows good prediction performance for both positive and negative site predictions for each of the three residues. It is also evident that the performance increases if more features are included in each training instance (i.e., increasing the window size).

ISSN: 1471-2105