Abstract
AbstractIdentifying high-risk individuals with diseases through reliable prediction models guides screening and preventive treatment. Most complex diseases have a genetic basis influenced by multiple genes and so disease risk can be estimated using polygenic risk score (PRS) algorithms. Many PRS algorithms have been developed so far. Among them, BayesR shows good characteristics of unbiasedness, accuracy, sparseness, and robustness. It detects the associated SNPs, estimates the SNP effects, and makes prediction of disease risks based on all SNPs simultaneously. However, this method assumes that the phenotypes follow a Gaussian distribution, which cannot be met in case-control studies. Here, we made an extension of the BayesR method, called BayesRB, by adding auxiliary variables to the BayesR model. We explored the characteristics, efficacy, and accuracy of BayesRB when estimating SNP effects and predicting disease risks compared with three traditional algorithms under different conditions using both simulated data and real data from the Welcome Trust Case Control Consortium (WTCCC). For SNP effect estimation, BayesRB shows unbiasedness and sparseness for big and small effect SNPs, respectively. For disease risk prediction, BayesRB had the best performance among the methods. This study provides a theoretical basis for complex disease risk prediction and disease prevention
Publisher
Cold Spring Harbor Laboratory