Heritability estimation using genetic similarity embedding
报告人：王健桥 (Harvard University)
时间：2024-02-29 14:00-15:00
地点：智华楼四元厅
Abstract:
We introduce a similarity-based framework for robust heritability estimation in genome-wide association studies (GWAS). Traditional fixed and random-effects models for heritability estimation often impose restrictive assumptions on regression coefficients or the design (genotype) matrix. These assumptions are usually violated in practice due to the heterogeneous genetic effects shaped by natural selection and the high genotype correlation caused by linkage disequilibrium. To overcome these limitations, we propose a SIMILarity Embedding method (SMILE) by modeling the relationship between the gram matrix of the outcome vector and the gram matrix of the underlying genetic signal vector, which correspond to the outcome similarity and genetic similarity respectively. SMILE includes the classical random-effects model as a special case and improves the fixed-effects model by not requiring accurate estimation of the inverse of the LD matrix or the regression coefficients. SMILE effectively captures heterogeneous genetic effects under weaker assumptions and makes the distinction between random-effect and fixed effect modeling approaches unnecessary. We develop a scalable implementation of SMILE that is capable of efficiently analyzing large biobank GWAS data.
Speaker:
Dr. Jianqiao Wang is currently a postdoctoral research fellow in the Biostatistics Department at Harvard T.H. Chan School of Public Health. He earned his Ph.D. in Biostatistics from the University of Pennsylvania in 2022. His research focuses on developing robust and scalable statistical methods for analyzing the high-dimensional genetic data.
