DeepWAS: Multivariate genotype-phenotype associations by directly integrating regulatory information using deep learning

Abstract

Genome-wide association studies (GWAS) identify genetic variants associated with quantitative traits or disease. Thus, GWAS never directly link variants to regulatory mechanisms, which, in turn, are typically inferred during post-hoc analyses. In parallel, a recent deep learning-based method allows for prediction of regulatory effects per variant on currently up to 1,000 cell type-specific chromatin features. We here describe "DeepWAS", a new approach that directly integrates predictions of these regulatory effects of single variants into a multivariate GWAS setting. As a result, single variants associated with a trait or disease are, by design, coupled to their impact on a chromatin feature in a cell type. Up to 40,000 regulatory single-nucleotide polymorphisms (SNPs) were associated with multiple sclerosis (MS, 4,888 cases and 10,395 controls), major depressive disorder (MDD, 1,475 cases and 2,144 controls), and height (5,974 individuals) to each identify 43-61 regulatory SNPs, called deepSNPs, which are shown to reach at least nominal significance in large GWAS. MS-and height-specific deepSNPs resided in active chromatin and introns, whereas MDD-specific deepSNPs located mostly to intragenic regions and repressive chromatin states. We found deepSNPs to be enriched in public or cohort-matched expression and methylation quantitative trait loci and demonstrate the potential of the DeepWAS method to directly generate testable functional hypotheses based on genotype data alone. DeepWAS is an innovative GWAS approach with the power to identify individual SNPs in non-coding regions with gene regulatory capacity with a joint contribution to disease risk. DeepWAS is available at https://github.com/cellmapslab/DeepWAS.

Identifying interpretable gene-biomarker associations with functionally informed kernel-based tests in 190,000 exomes

“…Deep-learning-derived functional annotations have been considered in other studies in the context of association testing. Proposed methods include signed LD-score regression [64], or the association tests presented in DeepWAS [65]. However, these methods have not been designed for rare variants.…”

The genetic basis of major depression

“…Several other associated loci have functions in neurite outgrowth, synaptic function and plasticity, as well as in immunity and inflammation (Wray et al, 2018). To establish the functions of the proposed risk SNPs (single nucleotide polymorphisms) and their links to specific genes, analyses of cell-type-specific gene expression and epigenetic data are necessary (Arloth et al, 2020; Chan et al, 2020; Li et al, 2020; Wang et al, 2020; Zhong et al, 2019).…”

GEPSi: A Python Library to Simulate GWAS Phenotype Data

“…Numerous computational methods have been developed to analyze Genome-Wide Association Study (GWAS) datasets, in order to identify germline genetic variants that are associated with phenotypic traits in a population. These include mixed linear models (Loh et al, 2015;Mbatchou et al, 2021), penalized logistic regression (Hoffman et al, 2013;Yang et al, 2020), decision trees (Botta et al, 2014;Schwarz et al, 2010), and neural networks (Arloth et al, 2020;van Hilten et al, 2020).…”