Abstract: Permutation tests are widely used in practice. However, these tests either need restrictive assumptions for validity, or are not applicable to high-dimensional data. This study considers permutation tests for high-dimensional mean comparisons. Here, in order to get around these restrictions, the test statistics are calculated based on pseudo samples generated using a "binning" procedure. The corresponding permutation tests are proved to be asymptotically consistent. We also consider a related problem for signal identification and establish the asymptotic properties of the tests. Simulation studies demonstrate the favorable performance of our methods compared with that of existing tests. Finally, the proposed method is applied to a genome-wide association study for seven complex human diseases to identify possible single nucleotide polymorphisms associated with the diseases.

Keywords: Consistency of test, high-dimensional data, permutation tests, signal identification, test of mean-difference.