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

Table 5

Accuracy of PSO-LocBact compared to other state-of-the-art methods on the well-known benchmark dataset S taken from [7, 9, 30].

Gram-negative bacterial proteins
Benchmark dataset S: predictor	Inner membrane (557 proteins)	Outer membrane (124 proteins)	Cytoplasm (410 proteins)	Extracellular region (133 proteins)	Periplasm (180 proteins)	Overall (1,404 proteins)
PSO-LocBact	547	116	387	129	171	1,350 (96.15%)
Gram-LocEN [25]	551	116	374	130	169	1,340 (95.44%)
PSORTb 3.0 [10]	529	114	380	117	168	1,308 (93.16%)
CELLO2GO [4]	519	107	383	128	170	1,307 (93.09%)
Gneg-PLoc [26]	454	68	362	59	87	1,030 (73.36%)
Gneg-mPLoc [7]	525	105	357	79	154	1,220 (86.89%)
iLoc-Gneg [24]	539	103	367	115	161	1,285 (91.52%)
Fuel-mLoc [23]	541	111	379	129	161	1,321 (94.09%)
Gram-positive bacterial proteins
Benchmark dataset S: predictor	Cell membrane (174 proteins)	Cell wall (18 proteins)	Cytoplasm (208 proteins)	Extracellular region (123 proteins)	Overall (523 proteins)
PSO-LocBact	174	18	206	122	520 (99.42%)
Gram-LocEN [25]	173	17	203	120	513 (98.08%)
PSORTb 3.0 [10]	169	14	203	112	498 (95.22%)
CELLO2GO [4]	149	10	197	121	477 (91.2%)
iLoc-Gpos [27]	167	12	198	110	487 (93.12%)
Fuel-mLoc [23]	170	17	202	117	506 (96.75%)
Gpos-PLoc [30]	—	—	—	—	379 (72.47%)
Gpos-mPLoc [9]	—	—	—	—	430 (82.22%)
ML-KNN [28]	—	—	—	—	78.71%
wML-KNN [29]	—	—	—	—	91.49%