SUMOhunt: Combining Spatial Staging between Lysine and SUMO with Random Forests to Predict SUMOylation

Table 4

Comparison of SUMOhunt with previous models/servers.

Predictor	AC	SN	SP	MCC	AUC	Training features	Evaluation process	+train/−train
SUMOplot	90%	80%	93%	0.48	—	Only sequence	Training	—/—
SUMOsp (2006); Thd: 18 and Thd: 4	92.71%; 80.43%	83.6%; 89.12%	93.08%; 80.07%	0.5012; 0.3232	0.73; —	Only sequence	5-fold and LOOCV	239+ve/—
Boshu Liu et al. (2007)	89.18%	—	—	—	—	Sequence + physicochemical properties	LOOCV	227+ve/226−ve
SUMOpre (2008)	97.71%	73.96%	97.67%	0.6364	0.87	Only sequence	5-fold CV	240+ve/6361−ve; Test: 28
SUMOsvm (2008)	97%	62%	99%	0.67	0.92	Sequence + solvent accessibility + secondary structure + evolutionary profiles	5-fold CV	241+ve/ 5741−ve; Test: 27+ve/—
SSPFS (2009)	84.4%	—	—	—	—	Sequence + physicochemical properties	LOOCV	191+ve/954−ve; Test: 21+ve/106−ve
SUMOtr (2010)	85%	95%	75%	0.68	0.85	Sequence + 3D structure + hydrophobicity	5-fold CV	57+ve/711−ve
SUMOhunt	97.56%	100	94	0.95	1	Sequence + several physicochemical properties	Training + test; 10-fold CV	270+ve/273−ve; Test: 23+ve/18−ve