Methods for Enhancing Polygenic Risk Prediction Models for Complex Disease

PROJECT SUMMARY Early screening and prevention of individuals at risk of complex diseases are important strategies for reducing morbidity and mortality. Polygenic risk scores (PRS) are the cumulative, mathematical aggregation of risk derived from the contributions of many DNA variants across the genome. PRS are an emerging technology in the field of disease risk prediction and have been shown to be correlated with disease incidence. While PRS have shown great promise for complex diseases, current PRS models are overly simplistic and have limited predictive power and clinical utility. PRS do not account for the effects of rare genetic variants or other risk factors (clinical, environmental, social determinants of health) on disease risk. Rare variants generally have greater effects on disease risk due to selective pressure, but only a small number of individuals carry any single rare variant. The sparsity of rare variants makes it difficult to directly incorporate them into PRS. Additionally, while it is known that clinical, environmental, and social risk factors also influence risk, few analyses have successfully integrated PRS with these important non-genetic factors. To address this issue, we will develop novel translational informatics methods that integrate clinical, environmental, and genetic data to improve disease risk prediction. We will assess the clinical utility of these integrated risk prediction models using cardiovascular disease (CVD) to evaluate the potential for translation to clinical use. Based on the complexity of CVD, we hypothesize that a comprehensive range of risk factors along with rare variants need to be incorporated into PRS to improve the risk prediction and maximize the clinical utility of PRS for CVD. To achieve our goal, our specific aims are: 1) To develop novel methods that incorporate rare genetic variants into Polygenic Risk Scores (PRS); 2) To evaluate Integrated Risk Models that combine clinical, environmental, and social risk factors with PRS; 3) To develop and evaluate deep learning models integrating genetic, clinical, environmental, and social risk factors; 4) To translate our integrated models into the electronic health record (EHR). If these specific aims are achieved, we will have a set of integrated models that can be used in downstream clinical implementation programs to ultimately have a translational impact on disease treatment and prevention. Using these novel computational risk prediction models for precision health, along with our EHR integration approaches, will allow for the translation of integrated risk prediction into routine clinical care.