NEW YORK – A Chinese research team has outlined a functional annotation-informed approach for improving the accuracy of polygenic risk scores (PRS) for complex traits or conditions.
"Our findings illustrate the substantial enhancement in the predictive performance of PRSs achieved by OmniPRS compared to existing methods and demonstrate the potential clinical translation of OmniPRS," senior and corresponding author Xingjie Hao, a researcher at Huazhong University of Science and Technology, and his colleagues wrote in a paper published in Cell Genomics on Tuesday.
OmniPRS is a scalable, biobank-level framework to assemble summary statistic and tissue-specific and non-tissue-specific functional annotation data from prior genome-wide association studies and research databases to come up with PRS showing improved prediction accuracy compared to those constructed primarily based on summary statistic data alone.
"The aggregated polygenic score combines individual polygenic scores from multiple functional categories, providing a comprehensive measure of the overall genetic risk for a disease or trait in an individual," the authors explained, noting that OmniPRS "distinguishes itself from other PRS methods that use functional annotation by quantifying the specific variance across different annotations and employing an integrated approach to better account for their differences and relationships."
After validating the OmniPRS framework in 135 simulations, the team used the approach to establish polygenic scores for 11 traits — ranging from height, body mass index, blood lipid measurements, and hypertension to type 2 diabetes, Alzheimer's disease, and psychiatric conditions.
With the help of data from the UK Biobank project, European participants in the 1,000 Genomes Project, and published GWAS signals and summary statistics, the investigators used the OmniPRS method to quickly establish functional annotation-informed polygenic scores that appeared to perform better than existing methods for predicting traits or conditions in real or simulated datasets, the researchers reported.
When it came to quantitative traits such as height, BMI, low-density lipoproteins, high-density lipoproteins, triglycerides, or total cholesterol, PRS established with the OmniPRS framework had prediction accuracies that improved some 52.3 percent, on average, compared to polygenic scores established with a so-called "clumping and thresholding" method. PRS accuracies increased some 8.4 percent for quantitative traits assessed by OmniPRS compared to PRS based on a Bayesian framework method.
Likewise, the team saw improved predictive performance of OmniPRS scores for binary traits such as the presence or absence of Alzheimer's disease, schizophrenia, bipolar disorder, hypertension, or type 2 diabetes.
Although they noted that the approach is expected to improve as additional functional annotation and variant data become available to assess models trained with ever-larger cohorts, the investigators suggested that the OmniPRS strategy "has the potential to become a widely adopted tool for assessing genetic contributions to risk prediction in clinical applications, particularly within biobank-scale datasets."
"Our findings unequivocally demonstrated and highlighted that leveraging multiple functional annotation categories could significantly improve the accuracy of polygenic prediction," the authors reported. "Such accurate genetic predictions of complex traits hold potential for advancing disease screening, improving early interventions, and supporting the development of personalized medicine."