Kernel-Based Aggregation of Marker-Level Genetic Association Tests Involving Copy-Number Variation

Genetic association tests involving copy-number variants (CNVs) are complicated by the fact that CNVs span multiple markers at which measurements are taken. The power of an association test at a single marker is typically low, and it is desirable to pool information across the markers spanned by the CNV. However, CNV boundaries are not known in advance, and the best way to proceed with this pooling is unclear. In this article, we propose a kernel-based method for aggregation of marker-level tests and explore several aspects of its implementation. In addition, we explore some of the theoretical aspects of marker-level test aggregation, proposing a permutation-based approach that preserves the family-wise error rate of the testing procedure, while demonstrating that several simpler alternatives fail to do so. The empirical power of the approach is studied in a number of simulations constructed from real data involving a pharmacogenomic study of gemcitabine and compares favorably with several competing approaches.