A unified Bayesian approach to transcriptome-wide association study

Transcriptome-wide association study (TWAS) has shed light on molecular mechanisms by examining the roles of genes in complex disease etiology. TWAS facilitates gene expression mapping studies based on a reference panel of transcriptomic data to build a prediction model to identify expression quantitative loci (eQTLs) affecting gene expressions. These eQTLs leverage the construction of genetically regulated gene expression (GReX) in the GWAS data and a test between imputed GReX and the trait indicates gene-trait association. Such a two-step approach ignores the uncertainty of the predicted expression and can lead to reduced inference accuracy, e.g., inflated type-I error in TWAS. To circumvent a two-step approach, we develop a unified Bayesian method for TWAS, combining the two datasets simultaneously. We consider the horseshoe prior in the transcriptome data while modeling the relationship between the gene expression and local SNPs and the spike and slab prior while testing for an association between the GReX and the trait. We extend our approach to conducting a multi-ancestry TWAS, focusing on discovering genes that affect the trait in all ancestries. We have shown through simulation that our method gives better estimation accuracy for GReX effect size than other methods. In real data, applying our method to the GEUVADIS expression study and the GWAS data from the UK Biobank revealed several novel genes associated with the trait body mass index (BMI).