Effect of adding fold-change criteria to significance testing of microarray data

Microarray experiments involve simultaneous testing of thousands of genes, thus leading to a large number of false-significant results. To combat this, a wide range of multiple-testing procedures have been developed to reduce false significance rates and related errors. In contrast, clinical investigators may choose alternative routes having intuitive appeal but less optimal statistical properties, such as requiring a 2-fold change. While the corresponding limitations are commonly recognized, the impact of such approaches has not been specifically characterized for practical scenarios. This study investigates the effect of combining the t-test (with the Benjamini-Hochberg adjustment) and 2-fold-change cut-off on the resulting levels of significance and power. Findings illustrate that both significance and power are often dominated by the 2-fold criteria, essentially negating the properties of the multiple comparisons adjustment. Other cases lead to more complex and potentially counter-intuitive results.