PSO-LocBact: A Consensus Method for Optimizing Multiple Classifier Results for Predicting the Subcellular Localization of Bacterial Proteins

Table 3

Accuracy of various consensus methods on the test sets.

Gram-negative bacterial proteins
Location:	Extracellular region (%)	Outer membrane (%)	Periplasm (%)	Inner membrane (%)	Cytoplasm (%)	Overall (%)
Single predictors (as shown in Table 2)	0–100	0–100	1.16–88.37	0–100	0–100	20.23–97.44
Consensus classifier: PSO-LocBact	100	100	94.19	100	100	98.84
Consensus classifier: majority voting	97.67	100	95.35	100	98.84	98.37
Consensus classifier: Naïve Bayes	100	98.84	94.18	100	98.84	98.37
Consensus classifier: logistic regression	98.84	100	97.67	95.35	98.84	98.14
Consensus classifier: average probability voting	98.84	100	90.69	98.84	98.84	97.44
Single predictor: FUEL-mLoc (2017)	79.07	97.67	96.51	93.02	82.56	89.76
Gram-positive bacterial proteins
Location:	Extracellular region (%)	Cell wall (%)	Inner membrane (%)	Cytoplasm (%)	Overall (%)
Single predictors (as shown in Table 2)	34.21–97.39	0–93.50	77.21–100	97.50–100	59.81–97.10
Consensus classifier: PSO-LocBact	97.39	94.80	100	100	98.07
Consensus classifier: majority voting	93.42	93.50	100	100	96.78
Consensus classifier: Naïve Bayes	69.73	92.20	100	100	90.67
Consensus classifier: logistic regression	89.47	100	100	100	97.43
Consensus classifier: average probability voting	96.05	87.01	100	98.73	95.49
Single predictor: FUEL-mLoc (2017)	86.84	81.82	100	100	92.28