Table 1.
Results on cancer gene expression data
| Data set (ref.) | n | MCF,% | CV,% | TA,%(ref.) | BS | BS0 | RFE | RFE0 | DE |
| Golub (2) | 72 | 32 | 97.0 ± 4.2 | 99.3 (28) | 537 | 0 | 35 | 154 | 1007 |
| Singh (4) | 136 | 43 | 92.6 ± 3.0 | 81.1 (27) | 99 | 0 | 48 | 312 | 3807 |
| Alon (1) | 62 | 35 | 81 ± 7.2 | 97.9 (29) | 19 | 0 | 55 | 94 | 303 |
| Wang (6) | 286 | 37 | 65 ± 4.3 | N/A | 0 | 0 | 261 | 1250 | 106 |
| van't Veer (5) | 97 | 47 | 62 ± 8.4 | N/A | 0 | 0 | 42 | 153 | 1 |
Results are ordered by prediction accuracy. n, number of samples; MCF, minority class frequency; CVA, balanced cross-validated prediction accuracy, mean ± std.dev.; TA, balanced prediction accuracy of bootstrap signature on an independent test set (reference given in parentheses); BS, significant genes using the bootstrap with SVM at 5% FDR; RFE, genes chosen by recursive feature elimination; BS0 and RFE0, gene chosen by the bootstrap and RFE methods respectively on randomized data. DE, differentially expressed genes using the t-test at 5% FDR.