Robust Inference With Ghostknockoffs in Genome-Wide Association Studies With Sample Relatedness

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Genome-wide association studies (GWASs) are crucial for understanding complex trait genetics.
• Knockoff-based methods show promise for identifying unique, potentially causal variants.
• The application of these methods in studies with related individuals, like the UK Biobank, is not well-established.

Purpose of the Study:

• To evaluate the statistical validity and effectiveness of integrating GhostKnockoffs with marginal association tests in GWASs involving related individuals.
• To extend the applicability of GhostKnockoffs to diverse GWAS settings, including meta-analyses with sample overlap and independent samples.

Main Methods:

• Integration of the GhostKnockoffs method with state-of-the-art marginal association tests.
• Evaluation using simulation studies to assess robustness and validity.
• Application to a meta-analysis of nine European ancestral GWASs and sequencing studies for Alzheimer's disease.

Main Results:

• The proposed analytical strategy demonstrates robustness to arbitrary relatedness structures when using generalized linear mixed models.
• The method's validity extends to GWAS meta-analyses with sample overlap under specific Z-score calibration.
• The approach is also effective for association test statistics beyond score tests in independent samples.

Conclusions:

• The integrated GhostKnockoffs and marginal association testing strategy is a valid and effective approach for genetic studies with related individuals.
• This method enhances the utility of knockoff-based approaches across various GWAS scenarios, including complex meta-analyses.
• The practical utility is confirmed through successful application to Alzheimer's disease genetic studies.