Table 1.

Summary of the algorithms, training data sources and feature groups used by different phosphorylation site predictors on the independent test.

Predictors	Algorithms	Data sources	Feature groups	G/K*
GPS 3.0	Hierarchical clustering9	Phospho.ELM18, PhosphoBase49	Phosphorylation site peptide (PSP) sequence similarities21	G/K
MusiteDeep	Deep CNN, LSTM50	UniProt/Swiss-prot20, RegPhos51	One-of-K coding for 16 upstream and downstream residues	G/K
Musite 1.0	Ensemble learning52	Phospho.ELM, UniProt, PhosphoPep53, PhosphAt19	K-nearest neighbour (KNN) scores5, disorder states, and amino acid frequencies54	G/K
NetPhos 3.1	Neural networks	PhosphoBase	Convolutional sparse encoding55 of local sequence contexts	G/K
KinasePhos 2.0	SVM23	PhosphoBase, UniProt20	Local sequence patterns and local coupling patterns8	K
PhosphoPredict	Random forest	Phospho.ELM	Amino acid type, PSS, DISO, solvent accessibility, and various protein functional features10	K
PhosphoPick	Bayesian networks56	Phospho.ELM, HPRD57	Protein-protein interactions58 and protein cell-cycle types59	K
PhosContext2vec	SVM	Phospho.ELM, UniProt	The Shannon entropy, the relative entropy, PSS, DISO, OP, ACH, and distributed contextual feature vectors	G/K

*G/K indicate General and Kinase-specific phosphorylation site prediction, respectively.