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

Marion Patxot,Julia Sidorenko,Zoltan Kutalik,Reedik Mägi,Lars Rönnegård,Daniel Trejo Banos,Athanasios Kousathanas,Alexander Holloway,Sven E Ojavee,Gerhard Moser,Matthew R Robinson,Etienne J Orliac,Peter M Visscher

doi:10.1038/s41467-021-27258-9

Abstract

We develop a Bayesian model (BayesRR-RC) that provides robust SNP-heritability estimation, an alternative to marker discovery, and accurate genomic prediction, taking 22 seconds per iteration to estimate 8.4 million SNP-effects and 78 SNP-heritability parameters in the UK Biobank. We find that only ≤10% of the genetic variation captured for height, body mass index, cardiovascular disease, and type 2 diabetes is attributable to proximal regulatory regions within 10kb upstream of genes, while 12-25% is attributed to coding regions, 32–44% to introns, and 22-28% to distal 10-500kb upstream regions. 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 of these four traits. Our open-source software (GMRM) provides a scalable alternative to current approaches for biobank data.

Highlights

We develop a Bayesian model (BayesRR-RC) that provides robust single nucleotide polymorphism (SNP)-heritability estimation, an alternative to marker discovery, and accurate genomic prediction, taking 22 seconds per iteration to estimate 8.4 million SNP-effects and 78 SNP-heritability parameters in the UK Biobank
We have genetic markers grouped into minor allele frequency (MAF)-LD-annotation specific sets, with independent hyper-parameters for the phenotypic variance attributable to each group, so that the mixture proportions, the variance explained by the SNP markers, and the mixture constants are all unique and independent across SNP marker groups
We find that 32–44% of the h2SNP is attributable to intronic regions, 12–25% is attributable to exonic regions, 22–28% is attributable to markers 10–500 kb upstream of genes, with proximal promotors, enhancers and transcription factor binding sites cumulatively contributing

Summary

Introduction

We develop a Bayesian model (BayesRR-RC) that provides robust SNP-heritability estimation, an alternative to marker discovery, and accurate genomic prediction, taking 22 seconds per iteration to estimate 8.4 million SNP-effects and 78 SNP-heritability parameters in the UK Biobank. As large-scale biobank data is increasingly available, methods that provide joint estimates of the marker effects in a single step by estimating the effect sizes as random under flexible prior formulations may become beneficial as they: (i) can account for differences in the variance contributed across MAF, LD or annotation groups providing unbiased MAF-LD annotation-specific genetic effect size estimates and h2SNP of different annotations, allowing for a contrasting of the genetic architectures of complex traits; (ii) give the probability that each marker, genomic region, annotation, genecoding region, or SNP is associated with a phenotype, alongside the proportion of phenotypic variation contributed by each, yielding test statistics that describe the gene architecture of complex traits and the uncertainty over the estimates; and (iii) provide improved genomic prediction, whilst providing a posterior predictive distribution for each individual. We validate our approach in large-scale simulation study and provide an empirical example using four traits measured in both the UK Biobank and Estonian Biobank data

Methods

Results

Conclusion