GARFIELD - GWAS Analysis of Regulatory or Functional Information Enrichment with LD correction

Abstract

Loci discovered by genome--wide association studies (GWAS) predominantly map outside protein--coding genes. The interpretation of functional consequences of non--coding variants can be greatly enhanced by catalogs of regulatory genomic regions in cell lines and primary tissues. However, robust and readily applicable methods are still lacking to systematically evaluate the contribution of these regions to genetic variation implicated in diseases or quantitative traits. Here we propose a novel approach that leverages GWAS findings with regulatory or functional annotations to classify features relevant to a phenotype of interest. Within our framework, we account for major sources of confounding that current methods do not offer. We further assess enrichment statistics for 27 GWAS traits within regulatory regions from the ENCODE and Roadmap projects. We characterise unique enrichment patterns for traits and annotations, driving novel biological insights. The method is implemented in standalone software and R package to facilitate its application by the research community. IntroductionGenome--wide association studies (GWAS) have discovered susceptibility variants for complex diseases and biomedical quantitative traits, with over 16 000 genotype--phenotype associations found to date 1,2 ,representing a large investment in resources, time and organisation to understanding human disease and other phenotypes. Despite the statistical soundness of the discovered associations, a large proportion (~90%) of implicated variants are classified as intronic or intergenic 3 and thus do not have a straightforward link to a cellular or molecular mechanism. This has prompted a number of efforts to annotate their putative functional . CC-BY-ND 4.0 International license not peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was . http://dx.doi.org/10.1101/085738 doi: bioRxiv preprint first posted online Nov. 7, 2016; consequences in cell specific contexts from experimentally derived regulatory genomic regions (e.g. regions marked by histone modifications, of open chromatin and transcription factor binding [3][4][5][6] ), principally as a means to inform and accelerate functional validation efforts.The robust identification of which combinations of cells and marks are most informative for a given disease or quantitative trait of interest (henceforth referred generically to as 'phenotype') requires that one can confidently identify biologically meaningful correlations. Genomic marks may cover a large proportion of the genome, and thus many disease--associated variants will be found within these marks by chance. In addition, the heterogeneous distribution of genetic variants and functional regions along the human genome, and thus non--random association with genomic features 7,8 , can create spurious correlations that again confound correct interpretation.Functional enrichment methods exploit experimentally derived regulatory genomic regions to assess the relative contributio...

Genetic Drivers of Epigenetic and Transcriptional Variation in Human Immune Cells

“…We focused on seven autoimmune diseases (celiac disease [CEL] [Dubois et al., 2010], inflammatory bowel disease [IBD] [Liu et al., 2015], including Crohn’s disease [CD] and ulcerative colitis [UC], multiple sclerosis [MS] [Beecham et al., 2013], type 1 diabetes [T1D] [Onengut-Gumuscu et al., 2015], and rheumatoid arthritis [RA] [Okada et al., 2014]), for which we retrieved publicly available genome-wide summary statistics. We first tested genome-wide enrichment for variants nominally associated with disease (p value ≤ 10 −5 ), applying an enrichment test controlling for LD, local gene density, and variant minor allele frequency (Iotchkova et al., 2016). We detected moderate-to-strong enrichment of disease associations for all classes of molecular QTLs tested and specific to autoimmune diseases with limited evidence for cell-type specificity of enrichments (Figure 7A).…”

“…In order to systematically measure the statistical significance of the overlaps between GWAS disease variants and molecular QTLs, we used GARFIELD (Iotchkova et al., 2016), a novel enrichment analysis approach taking genome-wide association summary statistics to calculate odds ratios for association between annotation overlap and disease status at given GWAS significance thresholds, while testing for significance via generalized linear modeling framework accounting for linkage disequilibrium, minor allele frequency, and local gene density. Linkage disequilibrium was calculated using SNPs from the combined UK10K and 1000 genomes Phase3 European cohorts.…”

Genome-wide association study of susceptibility to idiopathic pulmonary fibrosis

“…We used DeepSEA 31 , a deep-learning variant effect predictor, to identify whether any of the IPF risk signals were predicted to have a functional effect on chromatin features and transcription factor binding sites. The IPF risk signals were tested for enrichment in regulatory regions using FORGE 32 and GARFIELD 33 . SNPsea 34 was used to assess enrichment of genes in linkage disequilibrium with IPF risk variants in i) 1,751 genetic pathways and, ii) in genes showing differential expression between IPF cases and controls in four epithelial cell types 35 (Appendix).…”

An atlas of genetic variation for linking pathogen-induced cellular traits to human disease

“…H2P2 demonstrated enrichment of associated SNPs for 175 functional genome annotations. We used the GARFIELD package to calculate and visualize fold-enrichment of SNPs associated in H2P2 at variable p-value thresholds with different genomic features (50). In regards to SNP location, the greatest enrichment was observed for exonic SNPs ( Figure 2B).…”

“…Enrichment analyses were carried out using GARFIELD (50). GARFIELD has predefined a total 935 of 1005 features from ENCODE and the NIH Roadmap project, and applies generalized linear regression models while accounting for the effects of linkage disequilibrium (LD), minor allele frequency, and local gene density.…”