Genome-wide models for heritability and prediction
Supervisor: David Balding
Available for: PhD - preferred/MSc
Location: Melbourne Integrative Genomics, University of Melbourne
Project title: Genome-wide models for heritability and prediction
Description: The heritability of a phenotype is the fraction of its variance that can be modelled by genetics. "Genetics" for this purpose used to be measured by pedigree relatedness, but this has two major limitations: the result depends on the available pedigree, and the pedigree relatedness of two individuals can describe their expected genome sharing, but not the realised value. Nowadays, genome-wide allele sharing can be measured directly from SNP genotypes, but this has raised a lot of questions about what is the best way to represent the genetic similarity between two individuals, given their genome-wide genotypes. Analogy with the pedigree-based statistical model (a mixed regression model with variance matrix computed from pedigree-based kinship coefficients) has led researchers to a specific statistical model relating genome-wide SNP genotypes with a phenotype.
My work with collaborator Doug Speed (published recently in Nature Genetics, see below) has shown this model to be deficient. Based on a large-scale reanalysis of GWAS data for 43 phenotypes, we developed a model that better fits real data by taking into account the effects of minor allele fraction, linkage disequilibrium and genotype quality on the heritability of a SNP. Our new model leads to dramatic revisions of some published results in complex trait genomics.
This project will explore further implications of our superior heritability model. These include extending the model to multivariate phenotype analysis, and in particular to improved estimates of the genetic correlation between traits. Our LD model has substantial implications for LD Score Regression, a popular approach to analysing GWAS data that is available only in the form of summary statistics, rather than individual genotype data. In this project we will work both to further refine the new heritability model, including its extension to summary statistics, and to improve LD Score Regression and other related methods. This will in turn lead to better prediction models for individual and multiple phenotypes.
Speed D, …, Balding D (2017) Reevaluation of SNP heritability in complex human traits, Nature Genetics doi:10.1038/ng.3865