Table 1.
Comparison of performance of the different models for prediction of SUMOylation sites based on five-fold cross-validation *.
| Model | Sn | Sp | MCC | ACC | AUC |
|---|---|---|---|---|---|
| RF_AAindex [15] | 0.697 ± 0.004 | 0.650 ± 0.000 | 0.348 ± 0.004 | 0.674 ± 0.002 | 0.742 ± 0.003 |
| RF_BLOSUM62 [15] | 0.664 ± 0.004 | 0.650 ± 0.000 | 0.314 ± 0.004 | 0.657 ± 0.002 | 0.722 ± 0.003 |
| RF_EAAC [15] | 0.693 ± 0.008 | 0.650 ± 0.000 | 0.344 ± 0.008 | 0.672 ± 0.004 | 0.738 ± 0.005 |
| RF_ZScale [15] | 0.661 ± 0.006 | 0.650 ± 0.000 | 0.311 ± 0.006 | 0.655 ± 0.003 | 0.721 ± 0.004 |
| LGBM_AAindex [39] | 0.709 ± 0.011 | 0.650 ± 0.000 | 0.360 ± 0.011 | 0.680 ± 0.005 | 0.752 ± 0.003 |
| LGBM_BLOSUM62 [39] | 0.708 ± 0.013 | 0.650 ± 0.000 | 0.358 ± 0.013 | 0.679 ± 0.007 | 0.748 ± 0.005 |
| LGBM_EAAC [39] | 0.734 ± 0.008 | 0.650 ± 0.000 | 0.385 ± 0.008 | 0.692 ± 0.004 | 0.762 ± 0.004 |
| LGBM_ZScale [39] | 0.694 ± 0.012 | 0.650 ± 0.000 | 0.344 ± 0.012 | 0.672 ± 0.006 | 0.741 ± 0.005 |
| CNN_AAindex [31] | 0.784 ± 0.007 | 0.650 ± 0.000 | 0.438 ± 0.008 | 0.717 ± 0.004 | 0.788 ± 0.005 |
| CNN_BLOSUM62 [31] | 0.784 ± 0.006 | 0.650 ± 0.000 | 0.438 ± 0.007 | 0.717 ± 0.003 | 0.788 ± 0.004 |
| CNN_EAAC [31] | 0.771 ± 0.009 | 0.650 ± 0.000 | 0.424 ± 0.010 | 0.711 ± 0.004 | 0.782 ± 0.004 |
| CNN_ZScale [31] | 0.779 ± 0.008 | 0.650 ± 0.000 | 0.433 ± 0.009 | 0.714 ± 0.004 | 0.785 ± 0.005 |
| RSCNN_AAindex [63] | 0.803 ± 0.008 | 0.650 ± 0.000 | 0.458 ± 0.009 | 0.726 ± 0.004 | 0.799 ± 0.004 |
| RSCNN_BLOSUM62 [63] | 0.803 ± 0.007 | 0.650 ± 0.000 | 0.458 ± 0.008 | 0.726 ± 0.004 | 0.799 ± 0.003 |
| RSCNN_EAAC [63] | 0.763 ± 0.007 | 0.650 ± 0.000 | 0.416 ± 0.008 | 0.706 ± 0.004 | 0.774 ± 0.004 |
| RSCNN_ZScale [63] | 0.802 ± 0.007 | 0.650 ± 0.000 | 0.457 ± 0.007 | 0.726 ± 0.003 | 0.800 ± 0.004 |
* The model name is a combination of the algorithm and feature names. For example, RF_AAindex combines the RF algorithm and the AAindex feature. The abbreviations of the algorithms and features are described in “Materials and Methods”. Each measure (e.g., Sn, Sp, MCC, ACC, and AUC) is shown as the average ± standard deviation of the corresponding values of the models trained and evaluated based on five-fold cross-validation. The Sp values were fixed to allow fair comparison of the Sn, MCC, and ACC measures across different models.