Skip to content

Computing systematic polygenic risk scores associations using biobank-wide association scan

Yi Liu, Jie Zheng, Tom Richardson, Valeriia Haberland, Benjamin Elsworth, Matthew Lyon, Tom R Gaunt

We conduct a comprehensive and systematic risk associations among human traits and risk outcomes using polygenic risk scores (PRS) methods in order to identify and validate potential causal relationships in the human genome using hypothesis-free screening of risk factors from large scale genome-wide associations studies (GWAS). The IEU GWAS Database (Elsworth et al., fortcoming a) offers rich resources of GWAS across comprehensive phenotypes from large biobanks (such as the UK Biobank) accessible under a consistent format, and therefore allow for the analysis of systematic risk associations to dissect the complex relationships among phenotypic traits and diseases.

Traditionally PRS are calculated as the weighted sum of subjects' genome-wide genotypes at individual level, weighted by corresponding genotype effect sizes from summary statistic GWAS data. In our study, we adopt a novel approach (Yan et al., 2018 bioRxiv) based on which systematic PRS associations at biobank scale against target traits of interests can be computed at GWAS summary statistics level with accuracy comparable to the traditional PRS approach using individual level data (Richardson et al., 2018 bioRxiv), and thus we are able to carry out a large scale comprehensive association analysis between PRS traits and target traits from the studies of the IEU GWAS Database. Our results contribute to the integrative epidemiological evidence on EpiGraphDB ( where in a data-driven approach, hypothesis-free PRS associations across diverse traits would systematically identify putative causal associations where further rigorous investigations on causal factors can be achieved using Mendelian randomisation methods. In addition, we provide insights into the direct and indirect evidence of risk clustering across the broad spectrum of human traits by mapping the PRS associations with the semantic and phenotypic clusters of traits (Elsworth et al., forthcoming b).