Enhanced genetic fine mapping accuracy with Bayesian Linear Regression models in diverse genetic architectures

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Statistical genetic fine-mapping aims to identify causal variants influencing traits.
• Established methods like FINEMAP and SuSiE are widely used but have limitations.
• Bayesian Linear Regression (BLR) offers a flexible framework for genetic analysis.

Purpose of the Study:

• To evaluate Bayesian Linear Regression (BLR) models with BayesC and BayesR priors for statistical genetic fine-mapping.
• To compare the performance of BLR models against FINEMAP and SuSiE using simulations and real-world data.
• To assess the impact of genetic architecture on model performance.

Main Methods:

• Extensive simulations were conducted varying polygenicity, heritability, causal SNP proportion, and disease prevalence.
• Empirical analyses utilized UK Biobank (UKB) data, including over 6.6 million SNPs and 335,000 participants.
• Performance was measured using F1 classification scores and predictive accuracy.

Main Results:

• BLR models, especially with the BayesR prior, consistently outperformed FINEMAP and SuSiE in F1 scores.
• Predictive accuracy of BLR models was comparable to established methods.
• Region-wide BLR application generally improved F1 scores, except for highly polygenic traits.

Conclusions:

• BLR models, particularly BayesR, are accurate and robust tools for statistical genetic fine-mapping.
• These models provide valuable insights for both simulated and large-scale empirical genetic datasets.
• BLR offers a promising alternative for fine-mapping genetic associations.