Improved cytokine–receptor interaction prediction by exploiting the negative sample space

BMC Bioinformatics

Table 4 Performance metrics of different ML algorithms averaged over 10 × 10-fold cross validations

	SE	SP	ACC	MCC	AUC	g-means
Random (10 × 10-fold cross validation)
NB	51.04	61.22	55.49	0.106	0.560	66.2
A1DE	72.02	52.4	62.2	0.244	0.697	61.4
SMO-RBF	79.70	75.57	77.57	0.553	0.776	77.6
SMO-PolyK	64.72	49.45	57.11	0.144	0.571	56.5
SMO-PuK	71.76	83.29	78.53	0.555	0.775	77.3
IBK	81.14	71.43	76.28	0.528	0.762	76.1
Bagging	66.65	63.42	65.02	0.301	0.712	65.0
RF	72.77	69.79	71.27	0.426	0.795	71.2
K-means (10 × 10-fold cross validation)
NB	74.05	51.76	61.76	0.262	0.714	61.9
A1DE	83.70	85.02	84.34	0.688	0.901	84.4
SMO-RBF	88.55	90.48	89.53	0.796	0.903	89.5
SMO-PolyK	82.72	88.06	85.39	0.709	0.847	85.3
SMO-PuK	84.60	91.32	87.97	0.761	0.880	87.7
IBK	84.42	86.35	85.4	0.711	0.855	85.4
Bagging	86.10	92.25	89.19	0.786	0.938	89.1
RF	89.42	95.98	92.52	0.853	0.957	92.6

Each model was trained on different negative sample sets that have been generated using either random selection or the K-means based sampling

ISSN: 1471-2105