Extended Data Fig. 7: Association between genomic annotations and magnitude of allelic effects at rare variant associations.
From: Estimation and mapping of the missing heritability of human phenotypes
We quantified the association between the magnitude of allelic effects (hereafter denoted βj for variant j) and various genomic annotations using a linear regression model regressing \(\log ({\beta }_{j}^{2})\) onto \(\log [2{p}_{j}(1-{p}_{j})]\), with pj denoting the minor allele frequency of variant j, and indicators of annotation group (hereafter denoted, Aj). For each genomic annotation, variants without available information were grouped into an “other” category, to ensure comparable results across annotations. The y-axis in panel (a) represents the incremental R2 comparing the baseline model, \(\log ({\beta }_{j}^{2}) \sim \log [2{p}_{j}(1-{p}_{j})]\), with the full model containing the annotation, that is \(\log ({\beta }_{j}^{2}) \sim \log [2{p}_{j}(1-{p}_{j})]+{A}_{j}\). The y-axis in panel (b) represents the -log10 p-value from an analysis of variance comparing the full model with the baseline model. The red horizontal dotted line represents marginal significance, that is \({\log }_{10}(0.05)\approx 1.3.\) We highlight 4 genomic annotations with a marginally significant association with effect size magnitude: PrimateAI, “Functional”, WES-coverage and DCCVA. The PrimateAI annotation classifies variants with a normalized score below (resp. above) 0.4 as benign (resp. pathogenic). The “Functional” annotation has 6 groups: UTRs, Missense, Loff-of-Function, Intron, Synonymous and Intergenic. The WES-coverage annotation has 3 groups indicating variant outside of genomic regions covered by WES (WES loci), non-coding variants within WES loci, and coding variants within WES loci. The DCCVA (short for Distance to Closest Common Variant Association) was binned into 3 groups: less than 20 kb, between 20 kb and 100 kb, and more than 100 kb. Description of other genomic annotations is provided in the METHODS section.
