Reprioritizing genetic associations in hit regions using LASSO-based resample model averaging

Area of Science:

• Genetics
• Statistical Genetics
• Genomic Association Studies

Background:

• Single-SNP association testing is standard for identifying disease-related genomic regions.
• Challenges arise in complex regions with high linkage disequilibrium (LD), where multiple causal variants and their precise locations are ambiguous.
• Existing multi-SNP methods often lack systematic exploration of the model space and robust assessment of uncertainty.

Purpose of the Study:

• To develop a robust statistical method for reprioritizing single nucleotide polymorphisms (SNPs) within associated genomic regions.
• To address the ambiguity in identifying true causal variants in the presence of strong local LD.
• To provide a reliable way to characterize uncertainty in multi-SNP model selection.

Main Methods:

• Introduced LASSO local automatic regularization resample model averaging (LLARRMA), a novel approach combining LASSO shrinkage, resample model averaging, and multiple imputation.
• LLARRMA estimates the probability of each SNP's inclusion in a multi-SNP model across various data simulations.
• The method was validated using simulations based on case-control genome-wide association studies (GWAS) data.

Main Results:

• LLARRMA effectively identifies candidate SNPs in regions with multiple causal loci and strong LD.
• The method demonstrated superior performance in enriching true signals compared to single-locus analysis.
• LLARRMA outperformed Stability Selection, a recently proposed method, in simulation studies.

Conclusions:

• LLARRMA provides a powerful framework for navigating complex genetic architectures in association studies.
• The method enhances the ability to identify true causal variants in regions with high LD.
• LLARRMA offers a more reliable approach to multi-SNP model selection and uncertainty quantification in GWAS.