Table 1.
method | descriptor | mean AUC | sd | cv |
---|---|---|---|---|
RF | hydrophobicity | 0.927 | 0.001 | 0.001 |
molecular weight | 0.923 | 0.001 | 0.001 | |
IEP | 0.909 | 0.001 | 0.001 | |
pKa | 0.914 | 0.001 | 0.001 | |
cleavage site prediction | 0.851 | 0.003 | 0.003 | |
ANN | hydrophobicity | 0.841 | 0.028 | 0.034 |
molecular weight | 0.839 | 0.022 | 0.026 | |
IEP | 0.721 | 0.036 | 0.050 | |
pKa | 0.733 | 0.028 | 0.038 | |
cleavage site prediction | 0.762 | 0.036 | 0.047 | |
linear model | hydrophobicity | 0.826 | 0.008 | 0.009 |
molecular weight | 0.811 | 0.000 | 0.000 | |
IEP | 0.784 | 0.000 | 0.000 | |
pKa | 0.777 | 0.000 | 0.000 | |
cleavage site prediction | 0.803 | 0.000 | 0.000 | |
decision tree | hydrophobicity | 0.815 | 0.000 | 0.000 |
molecular weight | 0.841 | 0.000 | 0.000 | |
IEP | 0.771 | 0.000 | 0.000 | |
pKa | 0.764 | 0.000 | 0.000 | |
cleavage site prediction | 0.803 | 0.000 | 0.000 | |
JRip | hydrophobicity | 0.825 | 0.000 | 0.000 |
PART | hydrophobicity | 0.890 | 0.000 | 0.000 |
Rule372 | hydrophobicity | 0.710 | 0.000 | 0.000 |
Results of the 100-fold leave-one-out validation. The pro forma AUC values for the discrete methods (decision trees and rule based models) are just for comparison purposes. sd: standard deviation; cv: coefficient of variation.