Table 1.
The performance of DGM and DEG based hierarchy clustering and SVM classifiers on the TCGA ccRCC patient group and three independent ccRCC datasets.
| ccRCC patient cohorts | Normal | Tumor | Misclustered tissue samples |
AUC of the classifiers |
||
|---|---|---|---|---|---|---|
| DGM-based | DEG-based | DGM-based | DEG-based | |||
| TCGA-ccRCC | 72 | 539 | 2 | 4 | 0.942 | 0.767 |
| GSE36895 | 23 | 29 | 0 | 0 | 0.923 | 1.0 |
| GSE46699 | 63 | 67 | 9 | 15 | 0.953 | 0.949 |
| GSE40435 | 101 | 101 | 0 | 0 | 0.956 | 0.997 |
The DGM based classifier significantly outperformed the DGE based classifier by 22.8%((0.942 -0.767) / 0.767, Table 1 bold number) on the TCGA-ccRCC which is an imbalanced data set (72 normal vs 539 tumor samples).