Title: Covariate-modulated shrinkage estimator for imaging and genetics
Authors: Wesley Thompson - Institute of Biological Psychiatry (Denmark) [presenting]
Abstract: Estimates of Total Variance Explained, termed ``Heritability'' in Genetics applications, are useful in many contexts. In scenarios where the amount of variance explained by a set of predictors is substantial, but effects of interest are broadly distributed across a high-dimensional explanatory variables, it becomes useful to determine attributes or features of the variables that are ``enriched'' for strength of association. Features themselves may be high-dimensional, including multiple discrete and/or continuous variables. Moreover, predictors (e.g., SNPs or vertices) may be correlated (e.g., due to linkage disequilibrium or to spacial smoothness). We describe a novel Bayesian global-local shrinkage algorithm to estimate overall variance explained in this setting, which incorporates potentially high-dimensional covariates. The effects of the covariates are themselves regularized. The model is fitted via an MCMC algorithm. We apply this methodology to genetic and imaging data from the Adolescent Brain and Cognitive Development (ABCD) study, a 12,000 strong population study of US children aged 9-10 years at baseline, examining the association of whole-genome genotyping and brain imaging data on neurocognitive outcomes.