AbstractAssociation testing in genome-wide association studies (GWAS) is often performed at either the SNP level or the gene level. The two levels can bring different insights into disease mechanisms. In the present work, we provide a novel approach based on nonlinear post-selection inference to bridge the gap between them. Our approach selects, within a gene, the SNPs or LD blocks most associated with the phenotype, before testing their combined effect. Both the selection and the association testing are conducted nonlinearly. We apply our tool to the study of BMI and its variation in the UK BioBank. In this study, our approach outperformed other gene-level association testing tools, with the unique benefit of pinpointing the causal SNPs.
Nonlinear post-selection inference for genome-wide association studies
