Table 8.
Comparison of predictive accuracy of our method and MitPred for both plant and nonplant proteins in the TargetP data set
Predictor | Location | Sensitivity | Specificity | MCC |
Our method | plant mTP | 0.9348 | 0.8958 | 0.8587 |
MitPred (SVM) | plant mTP | 0.8234 | 0.8584 | 0.7418 |
MitPred (BLAST+SVM) | plant mTP | 0.9429 | 0.8422 | 0.8158 |
MitPred (HMM+SVM) | plant mTP | 0.8668 | 0.8484 | 0.7644 |
Our method | nonplant mTP | 0.7547 | 0.8333 | 0.7626 |
MitPred (SVM) | nonplant mTP | 0.8194 | 0.8863 | 0.8302 |
MitPred (BLAST+SVM) | nonplant mTP | 0.9380 | 0.9355 | 0.9268 |
MitPred (HMM+SVM) | nonplant mTP | 0.8571 | 0.9408 | 0.8830 |
Note: In our methods, parameters shown in Table 2 of the manuscript were used, but "posconstraint" and "negconstraint" were not specified for nonplant mTP. For MitPred, we used default parameters as follows. In SVM method, threshold was set as 0.5. E-value cutoff of BLAST+SVM method was 1e-4. E-value and SVM threshold for HMM based Pfam search+SVM method were 1e-5 and 0.5 respectively.