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

Area of Science:

• Genetics
• Statistical Genetics
• Computational Biology

Background:

• Identifying causal variants from genome-wide association studies (GWAS) is difficult due to variant sparsity and linkage disequilibrium.
• Accurate causal variant identification is crucial for understanding disease mechanisms and developing targeted therapies.

Purpose of the Study:

• Introduce BEATRICE, a novel framework for fine-mapping putative causal variants using GWAS summary statistics.
• Address challenges in causal variant identification, including sparsity and high correlation in genomic regions.

Main Methods:

• Employ a hierarchical Bayesian model with a binary concrete prior on causal variants.
• Utilize a variational algorithm minimizing KL divergence for fine-mapping.
• Integrate a deep neural network as an inference machine for parameter estimation.
• Implement stochastic optimization for sampling causal configurations and computing posterior inclusion probabilities.

Main Results:

• BEATRICE demonstrates uniformly better coverage, comparable power, and smaller credible set sizes than state-of-the-art methods.
• Performance gains increase with a higher number of causal variants.
• Successfully identified the APOE ε4 allele in an Alzheimer's disease GWAS, a feat not achieved by baseline methods.

Conclusions:

• BEATRICE is a valuable tool for identifying causal variants from GWAS and eQTL data.
• The framework enhances causal variant fine-mapping for complex diseases and traits.
• BEATRICE offers improved accuracy and robustness in identifying disease-associated genetic variants.