BG2: Bayesian variable selection in generalized linear mixed models with nonlocal priors for non-Gaussian GWAS data.

Area of Science:

• Genetics
• Statistical Genomics
• Bioinformatics

Background:

• Genome-wide association studies (GWAS) identify single nucleotide polymorphisms (SNPs) associated with phenotypes.
• Linear mixed models (LMMs) are common for GWAS but produce false discoveries and struggle with non-Gaussian data.
• Non-Gaussian phenotypes, like count data, require advanced analytical methods for accurate genetic association analysis.

Purpose of the Study:

• To develop a novel Bayesian method for identifying SNPs associated with non-Gaussian phenotypes in GWAS.
• To address the limitations of existing methods, particularly the high false discovery rate and inability to handle non-Gaussian data.
• To provide a flexible and accurate tool for analyzing complex genetic data.

Main Methods:

• Introduced Bayesian Generalized Linear Mixed Models for GWAS (BG2), a novel Bayesian approach.
• Utilized generalized linear mixed models (GLMMs) with novel nonlocal priors tailored for high-dimensional GWAS.
• Developed fast approximate Bayesian computations and a two-step procedure involving SNP screening and model selection.

Main Results:

• BG2 demonstrated favorable performance compared to GLMM-based single marker analysis (SMA) in simulation studies.
• The method effectively handles non-Gaussian phenotypes, including binary and count data.
• Case studies on cocaine dependence, alcohol consumption, and plant root development illustrated BG2's utility and flexibility.

Conclusions:

• BG2 offers a powerful and accurate Bayesian alternative for GWAS, especially for non-Gaussian phenotypes.
• The novel priors and computational methods enhance the analysis of complex genetic association data.
• BG2 provides a valuable tool for genetic research across various biological and medical applications.