Table 4.
Summarized best predictive performances for each classification task using RF model and three omics layers.
| Task | Method | MCC_cv (CI) | MCC_ts (CI) | PREC_cv (CI) | PREC_ts (CI) | REC_cv (CI) | REC_ts (CI) | Nf |
|---|---|---|---|---|---|---|---|---|
| BRCA-ER | juXT | 0.785 (0.776, 0.795) | 0.797 (0.778, 0.819) | 0.935 (0.932, 0.938) | 0.946 (0.935, 0.957) | 0.962 (0.959, 0.965) | 0.955 (0.949, 0.962) | 1801 |
| rSNF | 0.792 (0.782, 0.801) | 0.804 (0.779, 0.830) | 0.938 (0.935, 0.941) | 0.947 (0.934, 0.961) | 0.961 (0.958, 0.965) | 0.958 (0.949, 0.966) | 1801 | |
| rSNFi | 0.820 (0.808, 0.831) | 0.830 (0.803, 0.857) | 0.955 (0.951, 0.959) | 0.951 (0.939, 0.962) | 0.956 (0.952, 0.960) | 0.967 (0.956, 0.977) | 55.5 | |
| BRCA-subtypes | juXT | 0.778 (0.771, 0.785) | 0.795 (0.771, 0.817) | - | - | - | - | 1801 |
| rSNF | 0.769 (0.762, 0.777) | 0.811 (0.787, 0.835) | 1801 | |||||
| rSNFi | 0.788 (0.778, 0.798) | 0.838 (0.794, 0.879) | 301.5 | |||||
| KIRC-OS | juXT | 0.266 (0.243, 0.289) | 0.305 (0.229, 0.382) | 0.540 (0.509, 0.570) | 0.579 (0.494, 0.664) | 0.299 (0.280, 0.317) | 0.343 (0.300, 0.393) | 2319 |
| rSNF | 0.253 (0.230, 0.276) | 0.274 (0.189, 0.348) | 0.539 (0.505, 0.571) | 0.628 (0.507, 0.739) | 0.253 (0.235, 0.270) | 0.257 (0.200, 0.314) | 3313 | |
| rSNFi | 0.268 (0.239, 0.298) | 0.378 (0.288, 0.464) | 0.485 (0.449, 0.521) | 0.594 (0.512, 0.668) | 0.321 (0.296, 0.347) | 0.490 (0.380, 0.600) | 111 | |
| AML-OS | juXT | 0.141 (0.120, 0.163) | 0.223 (0.146, 0.307) | 0.675 (0.669, 0.681) | 0.704 (0.682, 0.725) | 0.860 (0.849, 0.870) | 0.880 (0.850, 0.907) | 6559 |
| rSNF | 0.180 (0.157, 0.202) | 0.263 (0.175, 0.366) | 0.685 (0.679, 0.691) | 0.717 (0.692, 0.743) | 0.876 (0.867, 0.886) | 0.873 (0.847, 0.903) | 656 | |
| rSNFi | 0.274 (0.245, 0.301) | 0.176 (0.068, 0.278) | 0.726 (0.718, 0.735) | 0.673 (0.639, 0.706) | 0.870 (0.858, 0.882) | 0.835 (0.785, 0.880) | 91.5 |
CI: 95% bootstrap confidence interval; {MCC,PREC,REC}_cv: best average MCC, precision, recall in cross-validation on training set splits; {MCC,PREC,REC}_ts: average MCC, precision, recall on test set splits; Nf: median number of features leading to MCC_cv. Bold indicates best performance (highest MCC and smallest signature size). Precision and recall were computed for binary classification tasks only.