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

BMC Bioinformatics

Table 8 Comparison of the proposed system with existing nine prediction systems in terms of Q₃ score according to Independent Benchmark.

Systems	Category	Prediction scores of the systems
		S (%)	T (%)	Y (%)
KinasePhos [11]		83.9	88.2	85.6
NetPhosK [10]	Kinase Specific	90.5	84.7	53.6
PPSP [6]		98.6	92.9	90.7
GPS [15]		17.5	16.5	14.4
AutoMotif Server AMS [14]		64.5	62.4	54.6
DISPHOS [8]		96.7	96.5	90.7
NetPhos [13]		16.6	22.4	16.5
PHOSIDA [5]	Kinase Independent	8.5	1.2	3.1
Scansite [12]		29.9	18.8	35.1
PPRED (Proposed system)		72.0	67.1	76.3

There are a total of 211 phosphorylated serine sites, 85 threonine sites and 97 tyrosine sites in the benchmark dataset. Here it can be found easily that the proposed system (PPRED) shows better accuracy than AutoMotif Server AMS, GPS, NetPhos, PHOSIDA and Scansite 2.0 for predicting S, T and Y sites.

ISSN: 1471-2105