Extended Data Fig. 3: Parameter estimation from SBayesRC and LDpred2 using summary statistics from a meta-analysis of two simulated cohorts. | Nature Genetics

Extended Data Fig. 3: Parameter estimation from SBayesRC and LDpred2 using summary statistics from a meta-analysis of two simulated cohorts.

From: Leveraging functional genomic annotations and genome coverage to improve polygenic prediction of complex traits within and between ancestries

Extended Data Fig. 3

The proportion of overlapped SNPs between the two cohorts varied from 100 to 0. When the proportion of overlapping is less than 100, there existed unequal per-SNP sample sizes in the GWAS summary data. a,b, SBayesRC gave approximately unbiased estimates for SNP-based heritability (true value = 0.5) and polygenicity (true value = 0.01), whereas these estimates in LDpred2 were largely biased. c, The model misspecification affected the residual variance in SBayesRC, which is a nuisance parameter in the model. Each box plot shows the spread of data in 10 independent simulations; the line is the middle (median), the box covers the middle half (IQR), the whiskers extend to 1.5 times the IQR, and dots show outliers. The dashed line indicates the true value in simulation.

Back to article page