Bayesian estimation of shared polygenicity identifies drug targets and repurposable medicines for human complex diseases.

Abstract

Background: Complex diseases may share portions of their polygenic architectures which can be leveraged to identify drug targets with low off-target potential or repurposable candidates. However, the literature lacks methods which can make these inferences at scale using publicly available data.

Methods: We introduce a Bayesian model to estimate the polygenic structure of a trait using only gene-based association test statistics from GWAS summary data and returns gene-level posterior risk probabilities (PRPs). PRPs were used to infer shared polygenicity between 496 trait pairs and we introduce measures that can prioritize drug targets with low off-target effects or drug repurposing potential.

Results: Across 32 traits, we estimated that 69.5 to 97.5% of disease-associated genes are shared between multiple traits, and the estimated number of druggable genes that were only associated with a single disease ranged from 1 (multiple sclerosis) to 59 (schizophrenia). Estimating the shared genetic architecture of ALS with all other traits identified the KIT gene as a potentially harmful drug target because of its deleterious association with triglycerides, but also identified TBK1 and SCN11B as putatively safer because of their non-association with any of the other 31 traits. We additionally found 21 genes which are candidate repourposable targets for Alzheimer's disease (AD) (e.g., PLEKHA1, PPIB ) and 5 for ALS (e.g., GAK, DGKQ ).

Conclusions: The sets of candidate drug targets which have limited off-target potential are generally smaller compared to the sets of pleiotropic and putatively repurposable drug targets, but both represent promising directions for future experimental studies.