Table 5.
Comparison of predictive performance on GDSC dataset for multiple drug sensitivity measures (AUC and 8 IC values) using both RF and FRF.
| Drug | Model | MAE | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| AUC | IC 10 | IC 20 | IC 30 | IC 40 | IC 50 | IC 60 | IC 70 | IC 80 | Mean | ||
| Erlotinib | RF | 0.0596 | 2.0831 | 1.7472 | 1.5039 | 1.3291 | 1.1948 | 1.0692 | 1.0133 | 1.0304 | 1.3714 |
| FRF | 0.0486 | 1.9813 | 1.6597 | 1.4382 | 1.2694 | 1.1357 | 1.0361 | 0.9867 | 1.0095 | 1.3146 | |
| Rapamycin | RF | 0.0640 | 4.3771 | 3.4771 | 2.9370 | 2.5294 | 2.2000 | 2.0355 | 2.0207 | 2.5359 | 2.7641 |
| FRF | 0.0636 | 4.3905 | 3.4525 | 2.8895 | 2.4642 | 2.1379 | 1.9446 | 2.0046 | 2.4707 | 2.7193 | |
| Sunitinib | RF | 0.0963 | 1.5494 | 1.5297 | 1.5542 | 1.6105 | 1.6518 | 1.7013 | 1.7750 | 1.8728 | 1.6556 |
| FRF | 0.0902 | 1.5306 | 1.5119 | 1.5378 | 1.5750 | 1.6276 | 1.6812 | 1.7428 | 1.8372 | 1.6305 | |
| PHA-665752 | RF | 0.0370 | 1.4403 | 1.2665 | 1.1492 | 1.0658 | 1.0002 | 0.9555 | 0.9539 | 0.9485 | 1.0975 |
| FRF | 0.0259 | 1.3522 | 1.2051 | 1.0999 | 1.0149 | 0.9546 | 0.9054 | 0.8954 | 0.9097 | 1.0422 | |
| MG-132 | RF | 0.1246 | 1.6207 | 1.6688 | 1.7445 | 1.7830 | 1.8549 | 1.9289 | 2.0313 | 2.1509 | 1.8479 |
| FRF | 0.1070 | 1.6062 | 1.6479 | 1.6968 | 1.7541 | 1.8117 | 1.8794 | 1.9619 | 2.0857 | 1.8055 | |
For FRF, node cost is calculated using 8 different IC regions. Bold values indicate the best performance.