Genome wide association studies enable the mapping of genetic effects to molecular phenotypes (such as gene expression levels or protein levels in specific tissues), and those genetic effects can then be mapped to disease traits for causal inference using a statistical framework known as Mendelian randomisation (e.g. Zheng et al 2021). However, recent work has demonstrated that the complexity of genetic relationships at the molecular phenotype layer far exceeds that which can be permitted by the assumptions used in previous causal modelling approaches (e.g. Min et al 2021). There are several issues that may be contributing to difficulties of causal modelling. First, our genetic effects on molecular phenotypes lack specificity because of e.g 1) genetic relationships being developed in readily available tissue samples which fail to separate differences in genetic effects between cell types within the tissue; or 2) the genetic effects are shared by different omic layers. Second, the standard approach to causal inference which treats a few molecules at a time in discrete time fails to capture the coordinated ‘state-based’ regulatory landscape within the cell.
Aims and objectives
- Develop cell-type specific genetic effect estimates on molecules by using emerging cellular deconvolution methods, especially in large scale datasets such as DNA methylation assays in the Million Veterans Program
- Generate an accurate genotype-phenotype map amongst molecular phenotypes within a single genomic region.
- Use heterogeneity of genetic effects across cell types and molecular phenotypes in a region to improve causal inference of the regional landscape on complex disease endpoints
- Evaluate through simulation whether the observed patterns of regional genotype-phenotype maps are consistent with state-based regulatory causal models
Methodology
This project will bring together large scale molecular and genetic datasets with an exploration of statistical methodology and simulation of biological models. The candidate will receive strong support from the supervisory team, which will also include researchers at Harvard who coordinate the Million Veterans Program.
Data sources:
- DNA methylation and protein measures on 50k samples in the Million Veterans Programme
- Protein levels on 50k samples in the UK Biobank
- Gene expression levels from GTEx and eQTLGen
- Genetic effects on disease endpoints from OpenGWAS (Elsworth et al 2020)
- MRC multi-ancestry cohort network
We will adapt methods such as SuSIE (Zoe et al 2021), PAINTOR (Kichaev et al 2017), hyprcoloc (Foley et al 2021), cellular deconvolution (Donovan et al 2020) and LD correction measures (Chen et al 2021) for construction of regional genotype-phenotype maps.
Simulations will be conducted using approaches such as those in the simulateGP package (https://github.com/explodecomputer/simulateGP/) as well us using dynamic single cell models (Pratapa et al 2020).
Causal models will build upon multivariable MR approaches (Sanderson et al 2018, Sadler et al 2021), and we will investigate the integration of structural priors and competing model selection approaches.
How to apply for this project
This project will be based in Bristol Medical School - Population Health Sciences in the Faculty of Health Sciences at the University of Bristol.
Please visit the Faculty of Health Sciences website for details of how to apply
Continue with Facebook
