Fig. 2
From: Non-parametric genetic prediction of complex traits with latent Dirichlet process regression models
Comparison of prediction performance of several methods with DPR.MCMC in simulations when PVE = 0.5. Performance is measured by R 2 difference with respect to DPR.MCMC, where a negative value (i.e., values below the red horizontal line) indicates worse performance than DPR.MCMC. The sample R 2 differences are obtained from 20 replicates in each scenario. Methods for comparison include BVSR (cyan), BayesR (chocolate), LMM (purple), MultiBLUP (green), DPR.VB (red), rjMCMC (black blue), and DPR.MCMC. Simulation scenarios include: a Scenario I, which satisfies the DPR modeling assumption; b Scenario II, which satisfies the BayesR modeling assumption; c Scenario III, where the number of SNPs in the large effect group is 10, 100, or 1000; and d Scenario IV, where the effect sizes are generated from either a normal distribution, a t-distribution or a Laplace distribution. For each box plot, the bottom and top of the box are the first and third quartiles, while the ends of whiskers represent either the lowest datum within 1.5 interquartile range of the lower quartile or the highest datum within 1.5 interquartile range of the upper quartile. For DPR.MCMC, the mean predictive R 2 in the test set and the standard deviation for the eight settings are, respectively, 0.272 (0.031), 0.299 (0.026), 0.295 (0.026), 0.281 (0.030), 0.277 (0.035), 0.278 (0.030), 0.282 (0.025), and 0.273 (0.022)
