Predicting gene regulation across populations to understand mechanisms underlying complex traits

点击翻译按钮获取中文摘要

Project Summary Most chronic diseases are polygenic with hundreds to thousands of causal variants, and we are starting to predict disease susceptibility with risk scores derived from genome-wide association studies. However, 77% of training data for these risk scores come from European ancestries populations and thus do not include genetic variants uniquely or more predominantly found in non-European populations, which limits both discovery and precision medicine potential. Methods that better identify causal variants and implicated biological mechanisms across populations are essential for equitable precision medicine implementation and can only be accomplished by studying the genetic architectures of complex traits in diverse populations. Since this project began, we have characterized the genetic architecture of the transcriptome and proteome within and across diverse populations. We identified a subset of transcripts and proteins that are well-predicted in one population, but poorly predicted in another and showed these differences are due, in part, to allele frequency and linkage disequilibrium differences. When testing prediction accuracy, we have shown that we need to consider both similarity in training and test population ancestries and total training sample size to optimally predict gene expression or protein abundance. In this proposal, we seek to drive mechanistic understanding of complex traits in diverse populations by (1) improving omics-trait prediction models for maximum utility within and between diverse populations and (2) investigating causal relationships between omics traits and complex traits and disease in diverse populations. We will integrate multi-omics data from African, African American, East Asian, European, and Hispanic populations in this project, including genome-wide genotype, transcriptome, proteome, and microbiome data. Since allele frequencies and linkage disequilibrium structures differ between populations due to different demographic histories, genetic prediction models trained in one population do not perform as well in another and thus are currently of limited utility for risk prediction and mechanistic interpretation. We will use fine-mapping, machine learning, and multivariate adaptive shrinkage to improve genotypic prediction of gene expression and protein levels across populations. Predicting the transcriptome and proteome from genotype data allows inference of whether high or low transcript or protein levels are associated with traits of interest, but false positives often result from linkage disequilibrium. We will integrate Mendelian randomization and colocalization sensitivity analyses into our PrediXcan method to test for causal relationships of transcripts, proteins, gut microbiota, or other exposures on disease outcomes across diverse populations. Together, our proposed aims have the potential to identify likely causal genes and molecular pathways underlying complex diseases. Our aims work toward development of effective risk assessment and potential treatment targets in diverse populations. Our team is well positioned to perform novel PrediXcan-based analyses of omics traits in diverse populations and promises to maximize impact by making our scripts, models, and results publicly available.