Probabilistic fine-mapping of transcriptome-wide association studies

Abstract

Transcriptome-wide association studies (TWAS) using predicted expression have identified thousands of genes whose locally-regulated expression is associated to complex traits and diseases. In this work, we show that linkage disequilibrium (LD) among SNPs induce significant gene-trait associations at non-causal genes as a function of the overlap between eQTL weights used in expression prediction. We introduce a probabilistic framework that models the induced correlation among TWAS signals to assign a probability for every gene in the risk region to explain the observed association signal. Our approach yields credible sets of genes containing the causal gene at a nominal confidence level (e.g., 90%) that can be used to prioritize and select genes for functional assays. Importantly, our approach remains accurate when expression data for causal genes are not available in the casual tissue by leveraging expression prediction from other tissues. We illustrate our approach using an integrative analysis of lipids traits where we correctly identify known causal genes.

Transcriptome-wide association studies: opportunities and challenges

“…S4). A pair of genes can have correlated predicted expression if a) the same causal eQTL regulates both genes, or b) two causal eQTLs in LD each regulate one of the genes 25 . Although only case a) counts as mechanistic co-regulation, we consider both cases together since they are not designed to be distinguishable by TWAS: the two genes' TWAS models can rely on distinct variants even in case a), or rely on the same variant even in case b).…”

“…To illustrate this phenomenon, we simulated expression and trait data (N trait = 10,000 individuals, N expression = 489 individuals from 1000 Genomes of European ancestry) for 1000 random genomic loci using the FOCUS simulation framework 25 (see "Suggested best practices and future opportunities"), and conducted TWAS using L2-penalized linear regression (see Methods). As expected, larger predicted expression correlation increased the probability of having a larger TWAS z-score than the causal gene (Table 2).…”

Genetic control of gene expression and splicing in the developing human brain

“…Similar to GWAS studies, TWAS genetrait association statistics in risk regions are correlated due to LD structure, and thus require fine-mapping for further interpretation. We applied FOCUS (Mancuso et al, 2017) to our fetalgene TWAS results that lie within a GWAS region to identify a credible set of causal genes at the SCZ GWAS significant loci (Methods) and identified a credible gene set consisting of 20 genes, including KCTD13 and NT5D2. KCTD13 and NT5D2 are both fetal specific hits in GWAS regions where adult specific hits also lie and show greater expression in prenatal time points compared to postnatal in the BrainSpan gene expression trajectories.…”

“…TWAS association statistics at genomic risk regions tend to be correlated as a function of linkage and eQTL overlap between predictive models (Mancuso et al, 2017;Wainberg et al, 2017). To prioritize candidate susceptibility genes in our TWAS we performed statistical finemapping using FOCUS (Mancuso et al, 2017).…”

“…TWAS association statistics at genomic risk regions tend to be correlated as a function of linkage and eQTL overlap between predictive models (Mancuso et al, 2017;Wainberg et al, 2017). To prioritize candidate susceptibility genes in our TWAS we performed statistical finemapping using FOCUS (Mancuso et al, 2017). FOCUS models the correlation structure induced by LD and overlapping eQTL weights across predictive models and computes posterior probabilities for a gene/intron to explain all observed TWAS association signal at a region.…”