LOS ANGELES – Researchers at Genomics PLC have developed a new set of polygenic risk scores for 53 diseases and traits using data from the UK Biobank.
At the American Society of Human Genetics annual meeting in Los Angeles last week, they presented an evaluation of the new risk scores, which are publicly available, using a benchmarking software tool that enables like-for-like performance evaluations of PRSs for the same disease or trait.
The new PRSs, detailed in a preprint published in MedRxiv in August, cover 28 diseases and 25 quantitative traits, split between a Standard PRS Set consisting of 28 diseases and eight quantitative traits and an Enhanced PRS Set containing all new PRSs.
"Our goal is to provide researchers with a state-of-the art, well-validated gold standard," said Vincent Plagnol, director of precision health at Genomics PLC, during a conference presentation.
Disease phenotypes in the dataset included prostate cancer, coronary artery disease, Crohn's disease, and schizophrenia; and traits included height, body mass index, age at menopause, and resting heart rate.
PRSs for the same trait in parents and their offspring showed utility in teasing apart genetic and environmental effects. While PRS-based mortality predictions between parents and offspring were generally highly correlated, stroke and coronary artery disease comprised two notable outliers wherein PRS appeared to play a much greater role in predicting parents' mortality than that of their children.
"We can speculate," Plagnol said, "that the environment has changed and mitigated the [impacts] of those diseases on mortality."
The software tool accompanying the PRS release addresses the challenge of comparing the performance of different PRS algorithms, enabling like-for-like comparisons of the predictive performance of different PRS scores in the biobank in an effort to correct for factors specific to cohorts and study designs, such as phenotype definition and other cohort characteristics.
The UK Biobank has a known bias toward individuals of European ancestry, and Plagnol commented that the new set of PRSs are more robust in people of similarly European background.
The Genomics PLC research team sought to account for this effect by maximizing non-European ancestry representation and reporting ancestry-specific results for all analyses with sufficient case data.
Although PRS predictive power declined in other ancestries relative to people of European ancestry, the predictive power of the UK Biobank Enhanced PRS in people of African ancestry appeared stronger for some diseases — such as type 2 diabetes and prostate cancer — than those for European ancestry individuals for other diseases, such as cardiovascular disease and age-related macular degeneration.
Plagnol also commented that his team observed similar PRS performance in a separate UK cohort from the 100,000 Genomes Project, and that performance across a set of non-UK cohorts remained "generally powerful," albeit more heterogeneous.
Nonetheless, caveats to the new PRSs include that despite the effort to include multiple ancestries, the representation of non-European groups in the UK Biobank is smaller than in other cohorts, and that small numbers of cases for rarer diseases precluded some comparisons.
Genomics PLC anticipates that further PRS releases and evaluation tools for other cohorts will enable more effective cross-cohort analyses going forward.
"We plan to update this dataset as new methods and data become available," Plagnol said.