Some statistical analyses on polygenic risk scores: genetic applications of random matrix theory
Bingxin Zhao, PhD Assistant Professor of Statistics and Data ScienceUniversity of Pennsylvania
Numerous statistical models have been proposed for genetic prediction using high-dimensional data from genome-wide association studies. The relative performance of these methods varies with the dataset characteristics and underlying genetic architecture of the targeted traits/diseases. Motivated by empirical observations, we present a series of analyses on genetic prediction methods in a high-dimensional sparsity-free setting, where we are allowed to have few to many true signals. In particular, I will present our analyses of two statistical problems. In the first project, we quantify the asymptotic bias when estimating genetic relationships with other traits based on polygenic risk scores ( Zhao, Yang, and Zhu, 2022, arXiv). The second project is to understand the impact of reference panels on genetic prediction accuracy ( Zhao, Zheng, and Zhu, 2022, arXiv). These analyses are based on recent advances in random matrix theory.
Hybrid: Meeting Room and Zoom Link to Follow 19 October 2022 1:30pm