|
Abstract:
|
It is often of interest to test whether a group of SNPs is associated with certain traits or diseases. For instance, natural groupings of SNPs may arise from their locations in a common gene or pathway of interest. Motivated by the Berk-Jones (BJ) statistic, which is known to have strong performance in rare-weak signal detection settings, we propose a new test for association between a SNP-set and an outcome - the generalized Berk-Jones (gBJ) statistic. The standard Berk-Jones statistic was constructed under the assumption that individual components of a set are independent. Our proposed generalized Berk-Jones statistics allow for an arbitrary correlation structure among SNPs, and gBJ is able to perform an accurate analytical p-value calculation accounting for correlation. We compare the performance of gBJ to other SNP-set tests across a range of genetic architectures, varying the signal strength, correlation structure, and sparsity of SNP-sets. We also apply the methods to analyze data from an infant neurodevelopment GWAS.
|
