A mixture model to decompose the heritable basis of complex traits

Thesis

Pharmacological drugs aim at modulating traits by targeting their causal mechanisms. However, knowledge of such causal mechanisms is scarce and constitutes the main bottleneck in drug development today. To address the issue, the consensus is to associate as many genetic variants as possible to traits of interest in order to then investigate their function and assign them to different mechanisms on that basis. With that purpose, there is an ongoing large-scale coordinated effort to systematically sequence and phenotype larger cohorts of individuals and to map functional elements in the genomes of human cells across tissues. Despite these advances, existing methods do not take full advantage of these increasingly available resources to associate genetic variants to traits and isolate the different mechanisms through which they operate. We have developed a mixture model that integrates multi-trait Genome-Wide Association Study (GWAS) z-scores and functional annotations of Single Nucleotide Polymorphisms (SNPs) to simultaneously boost GWAS power and group together SNPs that likely operate through a similar mechanism. The parameters of the model can be quickly inferred, and we show with realistic simulations that we can recover substantially more true associations than linear regression or a multi-trait GWAS meta-analysis method (MTAG), while recovering the simulated interpretable mixture model components. We applied our model to Coronary Artery Disease (CAD) and Autism Spectrum Disorder (ASD), finding three components for CAD (which point to the regulation of LDL, risk for smoking, and systolic blood pressure, respectively), and a single component for ASD that implicates the fetal and adult brain. Overall, our mixture model constitutes a powerful new framework to integrate the increasingly available functional annotations of the genome and multi-trait GWAS z-scores to uncover the mechanisms that drive complex traits. We expect our method to be progressively insightful as more and better data becomes available. This is expected to be specially true for complex neurodevelopmental disorders like ASD, whose driving mechanisms may only be detectable when considering developmental stage and cell-type specific functional annotations and traits.

Authors: Torada Aguilella, L

• DOI: 10.5287/ora-1zzkbbj19
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: GWAS; CAD; ASD; mixture model; heritability; functional annotations; biological pathways; genetic correlation
• Subjects: Genetics--Statistical methods