Probabilistic inference of the genetic architecture of functional enrichment of complex traits

Due to the complexity of linkage disequilibrium (LD) and gene regulation, understanding the genetic basis of common complex traits remains a major challenge. We develop a Bayesian model (BayesRR-RC) implemented in a hybrid-parallel algorithm that scales to whole-genome sequence data on many hundreds of thousands of individuals, taking 22 seconds per iteration to estimate the inclusion probabilities and effect sizes of 8.4 million markers and 78 SNP-heritability parameters in the UK Biobank. Unlike naive penalized regression or mixed-linear model approaches, BayesRR-RC accurately estimates annotation-specific genetic architecture, determines the underlying joint effect size distribution and provides a probabilistic determination of association within marker groups in a single step. Of the genetic variation captured for height, body mass index, cardiovascular disease, and type-2 diabetes in the UK Biobank, only [≤] 10% is attributable to proximal regulatory regions within 10kb upstream of genes, while 12-25% is attributed to coding regions, up to 40% to intronic regions, and 22-28% to distal 10-500kb upstream regions. [≥]60% of the variance contributed by these exonic, intronic and distal 10-500kb regions is underlain by many thousands of common variants, each with larger average effect sizes compared to the rest of the genome. We also find differences in the relationship between effect size and heterozygosity across annotation groups and across traits. Up to 24% of all cis and coding regions of each chromosome are associated with each trait, with over 3,100 independent exonic and intronic regions and over 5,400 independent regulatory regions having [≥]95% probability of contributing [≥]0.001% to the genetic variance for just these four traits. In the Estonian Biobank, we show improved prediction accuracy over other approaches and generate a posterior predictive distribution for each individual.