A stable and adaptive polygenic signal detection method based on repeated sample splitting
share
Using polygenic risk score for trait association analyses and disease prediction are paramount for genetic studies of complex traits. Valid inference relies on sample splitting, or more recently external data, to obtain a set of potentially associated genetic variants, along with their weights, for polygenic risk score construction. The use of external data has been popular, but recent work increasingly calls its use into question due to adverse effects of potential data heterogeneity between different samples. Our study here adheres to the original sampling-splitting principle but does so, repeatedly, to increase stability of our inference. To accommodate different polygenic structures, we develop an adaptive test for generalized linear models. We provide the asymptotic null distributions of the proposed test for both fixed and diverging number of variants. We also show the asymptotic properties of the proposed test under local alternatives, providing insights on why power gain attributed to variable selection and weighting can compensate for efficiency loss due to sample splitting. We support our analytical findings through extensive simulation studies and an application.
READ FULL TEXT
