Table 5 Overall performance metrics of classifier using data preprocessing combinations evaluated against ICGC/GEO test set related to Fig. 3
Index | Normalization | Batch effect correction | Scaling | Micro-average of AUROC | Weighted F1-score | p Value |
|---|---|---|---|---|---|---|
1 | Unnormalized | No batch correction | Unscaled | 0.95 (0.94–0.96) | 0.80 (0.76–0.81) | Baseline |
2 | Quantile normalization | No batch correction | Unscaled | 0.95 (0.91–0.97) | 0.75 (0.61–0.81) | 0.8986 |
3 | Quantile normalization with target | No batch correction | Unscaled | 0.96 (0.92–0.97) | 0.79 (0.64–0.85) | 0.7418 |
4 | Feature specific quantile normalization | No batch correction | Unscaled | 0.79 (0.79–0.80) | 0.50 (0.47–0.51) | 1 |
5 | Unnormalized | Batch correction | Unscaled | 0.87 (0.85–0.88) | 0.57 (0.55–0.60) | 1 |
6 | Quantile normalization | Batch correction | Unscaled | 0.86 (0.85–0.87) | 0.56 (0.55–0.62) | 1 |
7 | Quantile normalization with target | Batch correction | Unscaled | 0.87 (0.85–0.87) | 0.59 (0.56–0.65) | 0.9999 |
8 | Feature specific quantile normalization | Batch correction | Unscaled | 0.75 (0.74–0.76) | 0.22 (0.20–0.25) | 1 |
9 | Unnormalized | No batch correction | Scaled | 0.94 (0.92–0.95) | 0.65 (0.63–0.66) | 0.8986 |
10 | Quantile normalization | No batch correction | Scaled | 0.91 (0.87–0.93) | 0.62 (0.60–0.64) | 1 |
11 | Quantile normalization with target | No batch correction | Scaled | 0.90 (0.86–0.93) | 0.64 (0.63–0.66) | 1 |
12 | Feature specific quantile normalization | No batch correction | Scaled | 0.80 (0.79–0.82) | 0.53 (0.51–0.55) | 1 |
13 | Unnormalized | Batch correction | Scaled | 0.84 (0.81–0.87) | 0.57 (0.50–0.60) | 1 |
14 | Quantile normalization | Batch correction | Scaled | 0.86 (0.85–0.87) | 0.58 (0.54–0.62) | 1 |
15 | Quantile normalization with target | Batch correction | Scaled | 0.87 (0.85–0.88) | 0.59 (0.54–0.63) | 1 |
16 | Feature specific quantile normalization | Batch correction | Scaled | 0.77 (0.77–0.77) | 0.34 (0.30–0.39) | 1 |