Network diffusion-based risk scoring models for coronary artery diseases in UK biobank individuals:
Yonghyun Nam1§, Sang-Hyuk Jung1§, Dokyoon Kim1,2* 1 Department of Biostatistics, Epidemiology & Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA 19104, USA 2Institute for Biomedical Informatics, University of Pennsylvania, Philadelphia, PA 19104, USA
Poster # 1
Background: Intermediate clinical risk factors have been extensively studied for their interactions with complex diseases like coronary artery diseases. Composite polygenic risk scores (PRSs) have improved risk prediction accuracy by combining genetic predisposition of multiple intermediate risk factors. However, existing scoring models have limitations in leveraging extensive genetic relationships. To address this, we developed a network-based composite risk scoring model (net-cPRS) that incorporates genetic correlation networks. Methods: We constructed a genetic correlation network using linkage disequilibrium scores and GWAS summary statistics. This network captured the complex genetic relationships between the phenotype of interest (binary trait) and intermediate clinical factors. Network-based label propagations were then applied to estimate the polygenic impact on the phenotype by diffusing information to correlated clinical risk factors. These estimated impacts allowed us to combine multiple risk scores using regression models. Results: We tested net-cPRS using data from 370K European individuals in the UK Biobank for coronary artery diseases (CAD). Genetic correlation networks were built for CAD and 12 intermediate clinical factors. The network-diffused genetic impacts for CAD were obtained for each intermediate factor and used as weights in a logistic model. The scoring model was fine-tuned using individual genetic profiles from the UK Biobank. The results showed that net-cPRS achieved an average area under the curve (AUC) of 0.747, indicating improved prediction ability compared to standard PRS (AUC 0.65) and composite PRS (AUC 0.721). This proof-of-concept study highlights the potential of net-cPRS in enhancing risk stratification for complex diseases. Conclusion: Further experiments are necessary to apply net-cPRS to other phenotypes, and discussions are needed to determine the appropriate intermediate risk factors for scoring models.
LIGHTNING TALK - 2023 MidAtlantic Bioinformatics Conference