Stochastic model search with binary outcomes for genome-wide association studies

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWASs) generate vast amounts of data, necessitating advanced variable selection methods.
• The challenge of selecting relevant genetic predictors when the number of features vastly exceeds the number of binary outcomes is a significant hurdle in GWAS analysis.

Purpose of the Study:

• To introduce Binary Outcome Stochastic Search (BOSS), a novel Bayesian model search algorithm designed for variable selection in GWASs.
• To address the model selection problem in scenarios with a large number of genetic predictors and binary responses.

Main Methods:

• BOSS utilizes a latent variable model to connect observed outcomes with underlying genetic variables.
• A Markov Chain Monte Carlo approach is employed for efficient model searching and posterior probability evaluation of genetic predictors.

Main Results:

• Simulations demonstrate that BOSS achieves superior precision and good recall rates compared to stepwise regression, logistic lasso, and elastic net.
• BOSS successfully identified previously reported genetic polymorphisms associated with longevity and type 2 diabetes in real case studies.

Conclusions:

• BOSS represents a methodological advancement for model selection in high-dimensional genetic data.
• The algorithm effectively detects biologically relevant genetic markers while producing parsimonious models, outperforming existing techniques in both simulated and experimental data.