Significance of Gene Expression Changes Using the "Consistency" Test Peter Munson, Mathematical and Statistical Computing Lab, National Institutes of Health We have previously introduced the "consistency" test for determining the significance of expression changes in microarray experiments. This test shows substantially greater power than the paired t-test when the number of replicates is low. We investigated the behavior of this test its underlying assumptions have gained experience with several studies ongoing at the NIH. In most large-sample cases, results of the paired t-test and the consistency test coincide. For both tests, however, non-intuitive results can be obtained for a fraction of the tested genes. One requirement of the consistency test is that the error variance across genes is uniform. This can be easily satisfied using a suitable transformation. Two such transformations will be suggested. |
Graybill Conference |