Regarding parameter estimation, we found that GPA provided a satisfactory estimate of , the probability of being annotated for a certain group of SNPs, as long as there are enough SNPs in that group (Figures S32–S44 in Text S1). But we note that the estimates of the proportion of risk and non-risk SNPs, , may be biased (Figures S45–S66 in Text S1). This is no surprise because the distribution of the p-values of the risk SNPs obtained from the generative model in the simulation study may differ from the Beta distribution assumed in GPA. If the p-values of the risk SNPs are indeed generated from the Beta distribution, our GPA model can give fairly accurate estimates of (Figures S67-S72 in Text S1).
