The limits of predicting complex traits and diseases from genetic data

Results from genome-wide association studies (GWAS) have proven valuable for understanding the genetic architecture of complex traits and are potentially valuable for predicting disease risk. As GWAS sample sizes grow the prediction accuracy will increase and may eventually yield clinically actionable predictions, for example by stratifying individuals on risk. One limitation for making accurate disease risk prediction is the experimental sample size. We aim to quantify the limits of predicting disease risk for an individual by developing sophisticated statistical methods and applying them to quantitative traits in the large UK Biobank sample. Understanding of the limitations of predicting an individual?s risk of disease using genetic data is of great importance for disease prevention, and meets the UK Biobank?s stated purposes. Gaining accurate genetic risk predictors through the development of robust and powerful statistical methods, together with a large discovery sample (e.g. UK Biobank data), is critical for use in disease screening programs to stratify the population, which is expected to reduce the financial burden of the health system for the whole society. Through a focus on quantitative phenotypes, we will develop new approaches applicable to predicting disease risk. The genetic marker data will be used to estimate genome-wide relationships, which we will then correlate with phenotype. This analysis will simultaneously quantify how much of the observed individual differences in phenotype is due to genetic factors, and how accurate a genetic predictor can be. The accuracy of prediction will then be tested. We focus on well-characterised quantitative phenotypes of height, body mass index, blood pressure, osteoporosis, and metabolism. To have maximum power to predict risk of disease, we require access to the full cohort, because one of the main limiting factors of prediction is sample size. Our analyses will thus require individual-level imputed genotype and phenotype data. We request a wide range of phenotypes because prediction accuracy is sensitive to the underlying genetic architecture and we wish to quantify the limits of prediction across multiple diseases.