BEATRICE: Bayesian Fine-mapping from Summary Data using Deep Variational Inference

Abstract

AbstractWe introduce a novel framework BEATRICE to identify putative causal variants from GWAS summary statistics (https://github.com/sayangsep/Beatrice-Finemapping). Identifying causal variants is challenging due to their sparsity and to highly correlated variants in the nearby regions. To account for these challenges, our approach relies on a hierarchical Bayesian model that imposes a binary concrete prior on the set of causal variants. We derive a variational algorithm for this fine-mapping problem by minimizing the KL divergence between an approximate density and the posterior probability distribution of the causal configurations. Correspondingly, we use a deep neural network as an inference machine to estimate the parameters of our proposal distribution. Our stochastic optimization procedure allows us to simultaneously sample from the space of causal configurations. We use these samples to compute the posterior inclusion probabilities and determine credible sets for each causal variant. We conduct a detailed simulation study to quantify the performance of our framework across different numbers of causal variants and different noise paradigms, as defined by the relative genetic contributions of causal and non-causal variants. Using this simulated data, we perform a comparative analysis against two state-of-the-art baseline methods for fine-mapping. We demonstrate that BEATRICE achieves uniformly better coverage with comparable power and set sizes, and that the performance gain increases with the number of causal variants. Thus, BEATRICE is a valuable tool to identify causal variants from eQTL and GWAS summary statistics across complex diseases and traits.Author summaryFine-mapping provides a way to uncover genetic variants that causally affect some trait of interest. However, correct identification of the causal variants is challenging due to the correlation structure shared across variants. While current fine-mapping approaches take into account this correlation structure, they are often computationally intensive to run and cannot handle infinitesimal effects from non-causal variants. In this paper, we introduce BEATRICE, a novel framework for Bayesian fine-mapping from summary data. Our strategy is to impose a binary concrete prior over the causal configurations that can handle non-zero infinitesimal effects and to infer the posterior probabilities of the causal variant locations using deep variational inference. In a simulation study, we demonstrate that BEATRICE achieves comparable or better performance to the current fine-mapping methods across increasing numbers of causal variants and increasing noise, as determined by the polygenecity of the trait.