6.1 Multiple Comparisons

By default, the LIMMA wrappers from MSnSet.utils adjust the p-values to account for multiple comparisons using the Benjamini-Hochberg (BH) procedure. This controls the false discovery rate (FDR), and the resulting adjusted p-values are called q-values. To understand q-values, suppose we test whether there is a difference between the means of two groups. We do this for a set of 10,000 features. Suppose that, of these 10,000 tests, 1050 results in q-values less than \(\alpha = 0.05\) (a typical threshold for statistical significance). We would say that these 1050 most significant features have an estimated FDR of 0.05. That is, we expect that at most \(0.05 \cdot 1050 \approx 53\) of these results are false positives (features that are wrongly classified as significantly different between groups).