Table 2:
False-positive rates for the self-contained hypothesis for varying level of correlation in the gene set observed for different methods of gene set analysis
| Method | Correlation r of genes in G |
||||
|---|---|---|---|---|---|
| 0 | 0.2 | 0.4 | 0.6 | 0.8 | |
| Q1 | 0.05 | 0.096 | 0.178 | 0.22 | 0.282 |
| GSEA | 0.034 | 0.07 | 0.066 | 0.062 | 0.06 |
| GSA | 0.056 | 0.072 | 0.07 | 0.086 | 0.096 |
| GSA2 | 0.05 | 0.057 | 0.047 | 0.06 | 0.05 |
| SAFE | 0.053 | 0.043 | 0.06 | 0.04 | 0.057 |
| Globaltest | 0.05 | 0.036 | 0.038 | 0.062 | 0.066 |
| Q2 | 0.054 | 0.034 | 0.046 | 0.044 | 0.05 |
| ES.SC | 0.048 | 0.052 | 0.054 | 0.052 | 0.054 |
| ES.SC Analytic | 0.044 | 0.13 | 0.318 | 0.556 | 0.822 |
| PAGE | 0.03 | 0.148 | 0.32 | 0.606 | 0.702 |
| CATEGORY | 0.066 | 0.482 | 0.632 | 0.682 | 0.772 |
| PAR Wilcoxon | 0.056 | 0.516 | 0.644 | 0.716 | 0.802 |
Methods are denoted as in Table 1. The following specific settings of the methods were used: SAFE uses the Wilcoxon statistic; Globaltest uses the asymptotic null distribution; ES.SC uses null distribution based on permutation of samples; ES.SC Analytic uses the Kolmogorov–Smirnov null distribution; GSA uses all genes in the data set for re-standardization; GSA2 is the modified version of GSA with the P-value based on Equation (6). PAR Wilcoxon is the parametric method suggested by Irizarry et al. [16], see last row in Table 1.