Annotation-Informed Causal Mixture Modeling (AI-MiXeR) reveals phenotype-specific differences in polygenicity and effect size distribution across functional annotation categories

Abstract

1Determining the contribution of functional genetic categories is fundamental to understanding the 2 genetic etiology of complex human traits and diseases. Here we present Annotation Informed MiXeR: a 3 likelihood-based method to estimate the number of variants influencing a phenotype and their effect 4 sizes across different functional annotation categories of the genome using summary statistics from 5 genome-wide association studies. Applying the model to 11 complex phenotypes suggests diverse 6 patterns of functional category-specific genetic architectures across human diseases and traits. 7 8 Keywords: genetic architecture, complex traits, functional genetic categories, polygenicity, variant effect 9 size 10 11 Background 12 The rapid technological advances of the last years have provided an enormous amount of genetic data 13 thereby promoting the development of statistical methods aimed to unravel the genetic architecture of 14 complex traits [1]. A key effort has been to estimate SNP-based heritability, either using individual-level 15 genotype data [2], or summary-level statistics from genome-wide association studies (GWAS) [3]. 16 However, heritability estimates provide a limited picture of the genetic architecture underlying complex 17 phenotypes. For example, they are agnostic about the number of genetic variants influencing a 18 phenotype and their effect sizes [4]. Both of these quantities can vary and still result in the same 19heritability, which is proportional to their product [5,6]. 20Recently, we developed a model which allows the breakdown of SNP-heritability into the number of 21 variants influencing a given phenotype (non-null variants) and the distribution of their effect sizes using 22 summary statistics from GWAS and detailed population-specific linkage disequilibrium (LD) structure [5, 23 6]. This model assumes a uniform distribution of non-null variants with the same expected effect size 24 throughout the genome. However, prior genetic studies suggest that non-null variants are differentially for the same phenotype may differ depending on a GWAS's sample size, as well as on the coverage of 2 the tested variants. The samples of the GWAS tested here vary by more than one order of magnitude in 3 size, from approximately 5×10 4 for BD and ADHD to more than 7×10 5 for EA and height. In all 4 simulations, we kept the sample size constant (N = 10 5 ) and varied only the heritability (ℎ 2 =0.1, 0.4, 0.7). 5Since these quantities contribute to the GWAS z-scores distribution only through their product (follows 6 from formula (5)), our simulation scenario with N=10 5 and ℎ 2 =0.7 is equivalent to a scenario with, for 7 example, N=7×10 5 and ℎ 2 =0.1. Other aspects of potential GWAS-related issues (e.g. coverage of tested 8 Declarations 1

Linkage Disequilibrium and Heterozygosity Modulate the Genetic Architecture of Human Complex Phenotypes

“…Complex phenotypes are emergent phenomena arising from random mutations and selection pressure. Underlying causal variants come from multiple functional categories (Schork et al, 2013), and heritability is known to be enriched for some functional categories (Finucane et al, 2015; Gazal et al, 2017; Shadrin et al, 2019). Thus, it is likely that different variants will experience different evolutionary pressure either due to fitness directly or to pleiotropy with fitness related traits.…”

“…This research is facilitated by using new analytic approaches to interrogate structural features in the genome and their relationship to phenotypic expression. Some of these analyses take into account the fact that different classes of SNPs have different characteristics and play a multitude of roles (Schork et al, 2013; Finucane et al, 2015; Shadrin et al, 2019). Along with different causal roles for SNPs, which in itself would suggest differences in distributions of effect-sizes for different categories of causal SNPs, the effects of minor allele frequency (MAF) of the causal SNPs and their total correlation with neighboring SNPs are providing new insights into the action of selection on the genetic architecture of complex traits (Gazal et al, 2017; Wray et al, 2018; Zhang et al, 2018).…”

“…It is unclear how all these factors impact each other. A final and very important matter is that many analyses in the literature by other groups were conducted in the “infinitesimal” model framework (Finucane et al, 2015; Bulik-Sullivan et al, 2015; Gazal et al, 2017; Schoech et al, 2019; Speed and Balding, 2019), where all SNPs are causal, though there is compelling evidence that only a very small fraction of SNPs are in fact causal for any given phenotype (Zhu and Stephens, 2017; Zeng et al, 2018; Zhang et al, 2018; Frei et al, 2019; Shadrin et al, 2019; Holland et al, 2020).…”

The Oft-Overlooked Massively Parallel Reporter Assay: Where, When, and Which Psychiatric Genetic Variants are Functional?

“…These regions function to regulate transcript stability and miRNA interactions (16); they may also mediate nuclear retention/export of transcripts in the brain, which has only recently been explored (17). Coding variation carries a minority of heritable risk for schizophrenia, bipolar disorder, and attention deficit hyperactivity disorder (18). The occurrence of most diseaselinked variation in the least-well understood features of the genome/transcriptome thus obstructs understanding of disease biology.…”