Leveraging polygenic enrichments of gene features to predict genes underlying complex traits and diseases

Abstract

Genome-wide association studies (GWAS) are a valuable tool for understanding the biology of complex traits, but the associations found rarely point directly to causal genes. Here, we introduce a new method to identify the causal genes by integrating GWAS summary statistics with gene expression, biological pathway, and predicted protein-protein interaction data. We further propose an approach that effectively leverages both polygenic and locus-specific genetic signals by combining results across multiple gene prioritization methods, increasing confidence in prioritized genes. Using a large set of gold standard genes to evaluate our approach, we prioritize 8,402 unique gene-trait pairs with greater than 75% estimated precision across 113 complex traits and diseases, including known genes such as SORT1 for LDL cholesterol, SMIM1 for red blood cell count, and DRD2 for schizophrenia, as well as novel genes such as TTC39B for cholelithiasis. Our results demonstrate that a polygenic approach is a powerful tool for gene prioritization and, in combination with locus-specific signal, improves upon existing methods.

Genes with Specificity for Expression in the Round Cell Layer of the Growth Plate are Enriched in GWAS of Human Height

“…Although these partitioned regions do not directly parallel the zones of the post-natal growth plate, we were encouraged to find a trend congruent with the findings described here: regions representing earliest chondrocyte development (termed 1 and 2 by James et al) are significantly associated with height and remain significant after conditional analysis (Supplemental Table 6). Recently developed methods to prioritize genes from GWAS results offer the possibility to include many data sets simultaneously and identify those which are most helpful with gene prioritization [12]. Thus, the methodology described here provides an important technical resource for comparison of gene expression in different stages of chondrocyte development and will aid in the ongoing translation of GWAS associated SNPs to implicated causal genes for regulating skeletal growth.…”

Leveraging supervised learning for functionally-informed fine-mapping of cis-eQTLs identifies an additional 20,913 putative causal eQTLs

“…First, there are biases in the way the training variants are ascertained: the power to call a putative causal variant is affected by the recombination rate and the allele frequency of the variant 49,50 , and the GTEx cohort is highly biased towards adult samples with European ancestry background. Second, although we utilize over 6,000 features in EMS, larger sets of variant and gene annotations such as 3D configuration of genome 51,52 , constraint [53][54][55] or pathway enrichment 44 of genes could allow us to further improve prediction accuracy. Third, we simplified the prediction task by thresholding PIP.…”

“…We next compared the utility of PIPEMS to PIPunif for complex trait gene prioritization, as in Weeks et al 44 . To do this, we first calculated PIPEMS for 49 GTEx tissues using EMS of matched tissues as priors (Fig.…”

“…We then co-localized the eQTL signals with 95 UKBB phenotypes. Using the gold standard gene set described in ref [44], PIPEMS achieved higher precision and higher recall than PIPunif (Table. 1, Methods).…”

“…For complex trait causal non-coding variant prioritization, a threshold of PIP>0.1 was chosen to account for low sample size. We defined a gene prioritization task using 49 tissues in GTEx v8 and 95 complex traits in UKBB using the following steps (further details are described in Weeks et al 44 ):…”