Table 3.

Performance comparison of MU-LOC with existing tools for predicting plant mitochondrial proteins with N-terminal pre-sequences (independent testing set 2).

Tool	Parameter	Specificity	Sensitivity	Accuracy	Precision	MCC
MU-LOC(DNN)	Default	0.964	0.692	0.937	0.682	0.652
MU-LOC(SVM)	Default	0.974	0.662	0.943	0.741	0.669
TargetP	Sperschneider et al., 2017b	0.891	0.646	0.867	0.396	0.440
Predotar	Sperschneider et al., 2017b	0.944	0.600	0.910	0.542	0.520
YLoc	Sperschneider et al., 2017b	0.940	0.462	0.893	0.462	0.400
MitoProt II	Probability > 0.8a	0.842	0.600	0.817	0.295	0.329
MitoFates	Default	0.966	0.615	0.931	0.667	0.602
LOCALIZER	Sperschneider et al., 2017b	0.952	0.600	0.917	0.582	0.540

For the performance metrics used, the higher the value, the better the prediction accuracy. Results with the best performance are highlighted in bold. ^aThe webserver of MitoProt II only provides the prediction scores, and we chose a cut-off score of 0.8 to label the prediction class. ^bWe cited the performance metrics reported by Sperschneider et al. (2017) since the exact same testing set was used.