Eigenvalue significance testing for genetic association.

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Eigenvectors derived from genotype data are crucial for controlling spurious stratification in genetic association studies.
• Current significance testing relies on Tracy-Widom distribution, which assumes white-noise and can be inaccurate due to marker correlations and extreme eigenvalues.

Purpose of the Study:

• To develop and evaluate improved methods for determining null eigenvalue thresholds in genetic association studies.
• To address limitations of existing methods in accurately identifying true population stratification.

Main Methods:

• Introduced a novel block permutation approach to generate null eigenvalue distributions by preserving local genomic correlations.
• Proposed a fast, model-based approach using eigenvalue distribution modeling and the Marčenko-Pastur equation.
• Compared these methods against the standard Tracy-Widom approach and effective marker number calculations.

Main Results:

• Both block permutation and model-based approaches demonstrated strong performance in simulations and with 1000 Genomes project data.
• The standard method of using an 'effective' number of markers showed poor performance.
• The methods were validated using a cystic fibrosis genetic association study.

Conclusions:

• Novel block permutation and eigenvalue modeling methods provide more accurate null eigenvalue thresholds for genetic association studies.
• These improved methods are essential for reliable control of spurious stratification.
• The findings challenge the efficacy of traditional methods relying on effective marker counts.