Table 1.
Global scoring ROC AUC analysis to evaluate the ability of discriminating good and bad models in CASP13
| Method | GDT_TS | lDDT | CAD |
|---|---|---|---|
| Bhattacharya-ClustQ | 0.942 | 0.949 | 0.932 |
| FaeNNz | 0.889 | 0.947 | 0.937 |
| CPClab | 0.909 | 0.936 | 0.928 |
| MULTICOM_CONSTRUCT | 0.945 | 0.931 | 0.914 |
| MUfoldQA_T | 0.961 | 0.929 | 0.909 |
| ProQ3D-CAD | 0.880 | 0.926 | 0.930 |
| MUFoldQA_M | 0.959 | 0.922 | 0.900 |
| MULTICOM_CLUSTER | 0.937 | 0.922 | 0.904 |
| UOSHAN | 0.959 | 0.917 | 0.898 |
| ProQ4 | 0.875 | 0.913 | 0.907 |
Note: The data are extracted from http://predictioncenter.org/casp13/qa_aucmcc.cgi. The top 10 methods were selected according to a sorting by lDDT ROC AUC. ROC AUC values are displayed for lDDT and other scores relevant for CASP13 (GDT_TS, CAD). Single model methods (i.e. methods that need no ensemble of models as input) are highlighted with bold font, with FaeNNz (representing QMEANDisCo 3) conceptually being a quasi-single model method.