Bivariate logistic Bayesian LASSO for detecting rare haplotype association with two correlated phenotypes

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Joint modeling of correlated traits in genetic association studies enhances statistical power and reveals genetic etiology.
• Detecting rare genetic variants is crucial for understanding missing heritability.
• Existing methods like Logistic Bayesian LASSO (LBL) are limited to single binary phenotypes.

Purpose of the Study:

• To extend the Logistic Bayesian LASSO (LBL) method to handle multiple binary phenotypes.
• To develop a bivariate model that accounts for correlations between traits using a latent variable.
• To evaluate the performance of the bivariate LBL compared to the univariate LBL.

Main Methods:

• Development of a bivariate Logistic Bayesian LASSO (LBL) model incorporating a latent variable to link two binary outcomes.
• Extensive simulations were conducted to assess the performance of the bivariate LBL.
• Comparison of the bivariate LBL with the univariate LBL in various simulated scenarios.

Main Results:

• The bivariate LBL demonstrated comparable or superior performance to the univariate LBL across most simulation settings.
• The greatest power gain was observed when a haplotype influenced both traits, particularly when one effect opposed the trait correlation.
• Several associated rare haplotypes were identified in real datasets, including those for blood pressure and lung cancer/smoking.

Conclusions:

• The bivariate LBL is an effective extension for jointly analyzing multiple binary phenotypes in genetic association studies.
• This method enhances the detection of rare genetic variants contributing to complex diseases.
• The bivariate LBL provides a powerful tool for uncovering genetic etiologies underlying related traits.