Table 1.
Comparison of prediction accuracies achieved by RMA/RF and ANN (Ross et al., 2003).
| RMA/RF | RMA/RF‡ | Ross et al.† | |
| Analysis method | RF | RF | ANN |
| Samples in test set | 104 | 25 | 25 |
| BCR-ABL accuracy (samples) | 86.7 % (15)* | 89 % (4)* | 87.5 % (4)* |
| E2A-PBX1 accuracy (samples) | 100 % (18) | 100 % (5) | 100 % (5) |
| Hyperdip>50 accuracy (samples) | 100 % (17) | 99.8 % (4) | 95 % (4) |
| MLL accuracy (samples) | 100 % (20) | 100 % (5) | 100 % (5) |
| T-ALL accuracy (samples) | 100 % (14) | 100 % (2) | 100 % (2) |
| TEL-AML1 accuracy (samples) | 100 % (20) | 100 % (5) | 96 % (5) |
| Overall accuracy | 98.1 % | 98.2 % | 96.4 % |
*By cytogenetic analysis both cases that led to apparent inaccuracies are known to show the presence of BCR-ABL and >50 chromosomes.
‡ Average prediction accuracies from 100 independent analyses. For each analysis a new training and test set was chosen and discriminating probe sets were selected using the new training set.
† Average prediction accuracies from 10 independent analyses. For each analysis a new training and test set was chosen and discriminating probe sets were selected using the new training set.