p-values | Test | No point-mass | Dissonant | Consonant | Total | Validated | % Validated |
|---|
unadjusted | t-test | 3 | 0 | 245 | 248 | 63 | 25% |
| Â | Wilcoxon | 4 | 5 | 314 | 323 | 109 | 33% |
| Â | Two-part-t | 3 | 8 | 229 | 240 | 68 | 28% |
| Â | Two-part-W | 4 | 11 | 286 | 301 | 104 | 34% |
| Â | Empirical LRT | 4 | 7 | 271 | 282 | 81 | 28% |
BH-adjusted | t-test | 0 | 0 | 57 | 57 | 27 | 47.3% |
| Â | Wilcoxon | 3 | 1 | 137 | 141 | 58 | 41.1% |
| Â | Two-part-t | 0 | 3 | 66 | 69 | 30 | 43.4% |
| Â | Two-part-W | 3 | 6 | 103 | 112 | 55 | 49.1% |
| Â | Empirical LRT | 2 | 5 | 109 | 116 | 43 | 37.0% |
- The number of potential significant biomarkers when comparing the 67 cases and controls in the training set, based on unadjusted p-values (< 0.05) is shown for the t-test, WT, Two-part t-test, Two-part WT and empirical likelihood ratio test. In addition, the number of consonant, dissonant and no point-mass features among these is listed. All markers defined in the training set were investigated, aiming at validation, in an independent 2 × 67 test set. As is evident, the vast majority of potential biomarkers could not be validated. Lower panel: The number of potential significant biomarkers (p-values < 0.05) after BH adjustments for the t-test, WT, Two-part t-test, Two-part WT and empirical likelihood ratio test, and their performance in the independent test set are shown. While the expectation, that 95% of the potential biomarkers remain significant in the test set, could not be met, the percentage of biomarkers that could be validated is almost 2-fold in comparison to the unadjusted testing.
