A Repository of Polygenic Scores

An important output of genome-wide association studies are summary statistics from which it is possible to construct polygenic score (PGS): indices of SNPs that aggregate their explanatory power. Because PGSs are far more... [ view full abstract ]

An important output of genome-wide association studies are summary statistics from which it is possible to construct polygenic score (PGS): indices of SNPs that aggregate their explanatory power. Because PGSs are far more predictive of the phenotype than individual SNPs, they can be used in many analyses that would be underpowered if conducted using individual SNPs. For example, researchers have used polygenic scores as a measure of G in studies of gene-by-environment interactions. Going forward, sufficiently predictive polygenic scores may also prove valuable for targeting interventions (e.g., those most at risk of future Alzheimer’s disease) or as a control variable in evaluations of randomized experiments. The construction of PGSs with substantial predictive power has been enabled primarily by larger discovery samples. In the coming years, it is likely that the number of traits for which we can construct accurate PGSs will increase substantially. However, researchers who wish to use PGSs in their research continue to face a number of obstacles: (i) there are restrictions on the summary statistics that can be made publicly available and as a result, PGSs based on publicly available summary statistics are often less predictive than PGSs that could be constructed in principle (ii) from public summary statistics, it requires some effort to construct PGSs (especially for researchers doing this for the first time or researchers who wish to apply sophisticated methodologies), (iii) publicly available summary statistics may not be based on analyses of samples that are fully independent of the cohort for which the investigator wishes to construct polygenic scores, and (iv) because different researchers construct PGSs using different methodologies, it is hard to compare and interpret results from different studies. 

In this project, we aim to construct polygenic scores for a range of traits in some social-science data sets using a uniform methodology, and make the scores publicly available via mechanisms agreed upon by the cohort representatives. Since PGSs themselves are not SNP-level summary statistics, they can be disseminated more easily than the weights themselves. Directly sharing the polygenic scores (as opposed to just SNP-level summary statistics) may also contribute to progress by reducing barriers to entry for researchers interested in working with PGSs (especially researchers new to the field).