Evaluation of methodology for transcriptomic imputation into genome-wide association studies of complex human traits to infer causal mechanisms

This project will evaluate the performance of transcriptomic imputation methods for detecting association of gene expression with complex human traits. Genome-wide association studies (GWAS) have been successful at identifying loci associated with complex human traits. However, the biological mechanisms underpinning these associations remain poorly understood, and there has been little progress in the promised translation of GWAS findings into improved health outcomes. One approach to address this challenge is to apply causal modelling techniques, such as "Mendelian randomisation", to infer potential causal biological mechanisms through integrating GWAS data with transcriptomics in relevant tissues, using gene expression as an intermediate trait between DNA variation and phenotype. These techniques use genetic variation to examine the causal effect on disease of some modifiable "exposure" (such as gene expression) in the presence of potential confounding factors (such as lifestyle/environment). Transcripts associated with health-related outcomes may provide drug targets for novel interventions through down-regulation of the gene.

One disadvantage of Mendelian randomisation is the requirement for GWAS and transcriptomic data in the same individuals, which may be financially infeasible in large cohorts. A potential solution to this challenge is to use "two-sample" techniques that: (i) utilise external data resources (such as the Genotype-Tissue Expression "GTEx" Project) to identify subsets of genetic variants that predict tissue-specific gene expression; (ii) impute these transcriptomic profiles into existing GWAS using the identified multi-variant predictors; and (iii) test for association between imputed transcriptomic profiles and phenotype. The key advantages of this approach are that transcriptomic profiles can be imputed into large, extensively phenotyped GWAS at no cost (except computation), and "reverse causality" is not a major concern because phenotypic status or treatment does not alter germline genetic variation.

In this project, we will evaluate the performance of transcriptomic imputation methods for detecting association of gene expression with complex human traits. We will use computer simulations to compare the power of imputed transcriptomics with directly measured gene expression under a range of models of confounding with non-genetic factors. We will develop multi-variant predictors of gene expression in multiple tissues using available resources from GTEx, and use these to impute transcriptomic profiles into large-scale, deeply phenotyped GWAS available to the supervisory team, including UK Biobank (500,000 individuals), Estonian Biobank (50,000 individuals), and the Resource for Genetic Epidemiology Research on Ageing Cohort (100,000 individuals). As well as containing information on DNA variation (in the form of GWAS data and whole-genome sequence), these data sets also contain a variety of phenotypic measurements including data (measurements taken) on biological samples, additional exposure data not collected at the assessment visit (e.g., data from web-based dietary questionnaires), and data on health-related outcomes via linkage to a range of health-related records).

The output from the studentship will inform future design/analysis of GWAS, and will detect causal genes for complex human traits, thereby providing insight into regulatory mechanisms, and, for disease-related outcomes, identifying potential drug targets for novel therapeutic intervention.
The project maps to the BBSRC strategic research priorities "Bioscience for Health" and "World Class Underpinning Bioscience" by aiming to elucidate biological mechanisms underpinning complex phenotypic outcomes, including both long-term health-related outcomes and those related to normal physiological processes.