Fig. 2: Generative models of disease liability and various genetic architecture models.
From: Contextualizing genetic risk score for disease screening and rare variant discovery
We calculated the disease risk probability assuming two models of disease liability, namely, the liability-threshold model (green panel) and logit risk model (blue panel). To simulate the effect sizes of common causal variants (and the resulting phenotypes), we considered three genetic architecture models: polygenic, negative selection, and LD-adjusted kinship. For each model of disease liability and model of genetic architecture, we varied the simulation parameters (the white panels), including the heritability \({h}_{{{\mathrm{PB}}}}^{2}\,\)due to the common-variant polygenic burden (PB), heritability \({h}_{{{\mathrm{LEV}}}}^{2}\,\)captured by the (primarily rare) LEV and its allele frequency \(f\), disease prevalence K, and sample size N (as a study-design parameter). The chosen values are shown in the brackets under each parameter. In simulated data, we calculated the utility of the PB-LEV correlation. In addition, varying the proportion of noncausal variants (\({\pi }_{0}\)) in the estimate of PB, we quantified the power to detect the PB-LEV correlation. For each of the parameter combinations, we simulated 500 times to calculate the utility and power. These generative models of disease liability and genetic architectures provide the basis for a summary-statistics-based framework for inferences based on PB and LEV.
