TABLE 2.
Area under ROC curve of machine‐learning methods in predicting mutations and subtypes
| Characteristic | Logistic regression |
Random forest |
Support vector machine |
|---|---|---|---|
| Somatic mutation | |||
| TP53 | 0.715 | 0.930 | 0.885 |
| CDKN2A | 0.689 | 0.913 | 0.816 |
| NOTCH1 | 0.697 | 0.903 | 0.825 |
| NSD1 | 0.723 | 0.828 | 0.815 |
| PIK3CA | 0.726 | 0.871 | 0.827 |
| Transcriptional subtype | |||
| Basal | 0.726 | 0.954 | 0.845 |
| Mesenchymal | 0.783 | 0.930 | 0.864 |
| Atypical | 0.723 | 0.905 | 0.862 |
| Classical | 0.592 | 0.864 | 0.733 |
| Methylation subtype | |||
| Normal‐like | 0.680 | 0.942 | 0.835 |
| Hypo‐methylated | 0.781 | 0.881 | 0.814 |
| Hyper‐methylated | 0.740 | 0.968 | 0.859 |
| CpG island hyper‐methylated | 0.732 | 0.911 | 0.844 |