Incorporating substrate sequence motifs and spatial amino acid composition to identify kinase-specific phosphorylation sites on protein three-dimensional structures

Table 3 The comparison among PredPhospho, PPSP, GPS 2.0, KiasePhos 2.0, and our method.

Tools	PredPhospho	GPS 2.0	PPSP	KinasePhos 2.0	Our method
Method	SVM	GPS	BDT	SVM	SVM
Training feature	Sequence	Sequence	Sequence	Sequence	Sequence + 3D structural information
Material	PhosphoBase + Swiss-Prot	Phospho.ELM	Phospho.ELM	Phospho.ELM + UniProtKB	Phospho.ELM + UniProtKB
No. of kinase groups	4	> 100	68	58	> 100
Data input	Sequence	Sequence	Sequence	Sequence	Sequence, PDB ID or structure
3D structure visualization	-	-	-	-	JMol
PKA group	Sn = 70.1% Sp = 86.4%	Sn = 88.2% Sp = 86.6%	Sn = 86.9% Sp = 83.1%	Sn = 86.9% Sp = 85.6%	Sn = 89.4% Sp = 87.7%
PKC group	Sn = 70.9% Sp = 86.5%	Sn = 86.2% Sp = 83.0%	Sn = 82.9% Sp = 85.5%	Sn = 0.84 Sp = 0.86	Sn = 84.3% Sp = 89.1%
CK2 group	Sn = 82.0% Sp = 92.8%	Sn = 81.4% Sp = 86.4%	Sn = 84.0% Sp = 90.5%	Sn = 86.2% Sp = 86.4%	Sn = 88.1% Sp = 90.2%
SRC group	-	Sn = 82.3% Sp = 86.8%	Sn = 78.0% Sp = 74.6%	Sn = 86.4% Sp = 82.2%	Sn = 86.4% Sp = 86.2%

The highlights are marked in bold. For PKA group, our method has highest sensitivity and specificity. For PKC group, GPS 2.0 has highest sensitivity and our method has highest specificity. For CK2 group, our method has highest sensitivity and PredPhospho has highest specificity. For SRC group, our method has highest sensitivity and GPS 2.0 has highest specificity.
Abbreviation: SVM, support vector machine; MCL, Markov cluster algorithm; GPS, group-based phosphorylation scoring method; BDT, Bayesian decision theory; MDD, maximal dependence decomposition; HMM, hidden Markov model; AAC, amino acid composition; CP, coupling pattern; SA, structural alphabet; Sn, sensitivity; Sp, specificity; Acc, accuracy.