Comparison of mixed model based approaches for correcting for population substructure with application to extreme...

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Mixed models are essential for correcting confounding in genome-wide association studies (GWAS) due to population stratification and relatedness.
• Existing mixed model approaches have been validated for continuous and case-control traits but not specifically for extreme phenotype sampling (EPS).
• EPS involves collecting genetic data only from individuals with extreme trait values, posing unique analytical challenges.

Purpose of the Study:

• To compare the performance of established binary trait mixed model approaches (GMMAT, LEAP, CARAT) in the context of EPS data.
• To evaluate the utility of a linear mixed model implementation (GEMMA) for binary traits within an EPS framework.
• To assess type I error rates and statistical power across different mixed model methods under simulated population substructure.

Main Methods:

• Utilized simulation studies to estimate type I error rates and power for GMMAT, LEAP, CARAT, and GEMMA.
• Applied the evaluated methods to a real-world case-control dataset from Québec, Canada, known for its population substructure.
• Focused on assessing performance for both common and rare genetic variants.

Main Results:

• LEAP and GMMAT demonstrated effective control of type I error rates for common variants in EPS data, while CARAT showed inflated rates.
• Similar type I error control was observed when analyzing the real-world Québec dataset.
• For rare variants, the false positive rate remained elevated across methods, even after mixed model corrections.
• Methods that controlled type I error exhibited comparable statistical power.

Conclusions:

• The choice of mixed model methodology significantly impacts type I error control in EPS studies.
• Researchers must carefully select models that align with the specific sampling strategy (EPS) and the minor allele frequency of candidate single nucleotide polymorphisms (SNPs).
• Further investigation is needed for robust mixed model approaches for rare variants in EPS settings.