Enabling improved applicability and transferability of polygenic scores across populations

Polygenic scores – which quantify inherited risk by integrating information from many common sites of DNA variation – hold considerable promise for enabling a tailored approach to clinical medicine. However, alongside considerable (and warranted) enthusiasm, we and others have highlighted a crucial issue – current polygenic scores do not perform uniformly well across Americans. By assembling a team with deep expertise in statistical genetics, clinical informatics, data sharing, and genomic medicine, we outline the Functional and Fine-Mapping Approach to Improve Responsible Risk-modeling of Polygenic Risk Scores (‘FFAIRR-PRS’) approach to systematically address the key factors driving diminished performance. To enable analysis by the NHGRI consortium within the ANVIL ecosystem, we will contribute genetic and rich phenotype data across a better breadth of individuals to address these gaps. We leverage various cohorts and biobanks to patch together groups reflecting the broader U.S. population-at-large. This training data is needed to ensure that polygenic risk scores are as accurate as possible for all Americans. Individual-level data in ANVIL will be paired with curated summary association statistics and relevant annotations across ancestries, which will enable enhanced fine-mapping, sore weighting, and transethnic benchmarking activities. Our Study Site aims to (1) Aggregate and harmonize genotyping and phenotype data and deliver a sharable and scalable end-to-end analytic pipeline that starts with genotyping array data and a phenotype file and enables automated output of polygenic score benchmarking parameters; (2) Develop and share the new ‘FFAIRR-PRS’ statistical genetics framework, leveraging: (i) fine-mapping to assign causal probabilities based on >180 functional genomic annotations; (ii) incorporating correlations between effect sizes across traits; and (iii) integration of GWAS data across global populations reflecting the U.S. populace; and (3) Benchmark FFAIRR-PRS scores for 27 important phenotypes in multiple datasets, and develop risk models that integrate genetic and nongenetic factors. Performance will be benchmarked in accordance with ClinGen Complex Disease Working Group recommendations and compared across individuals reflective of the U.S. population. Beyond enhanced polygenic scores – aware of an ultimate aim of clinical implementation – we will develop a framework for integrated absolute risk models calibrated to the U.S. population that account for rare monogenic variants of large effect, family history, lifestyle, and clinical risk factors by adapting the Individualized Coherent Absolute Risk Estimator (iCARE) tool developed by co-I Chatterjee.