Abstract:
|
Rare variants are of increasing interest to genetic association studies largely due to their etiological contributions in human complex diseases. However, rare variants are difficult to detect individually due to the rarity of the mutant events. Collapsing analyses improve detection signal by aggregating information from multiple loci but are not able to pinpoint causal variants within a variant set. To perform inference at a localized level, additional information, e.g., on structure or predicted function, is needed to boost signal strength. We propose a rare variant association test which utilizes protein tertiary structure to increase signal and identify likely causal variants. Following the biological hypothesis that important variants are likely to cluster together in 3D protein space, we perform structure-guided collapsing, leading to local tests which borrow information from neighboring variants on a protein and provide association information on a variant-specific level. We use a kernel machine framework along with resampling to evaluate significance and show the utility of the proposed method using simulations and a real data application on the ACCORD clinical trials.
|