Fast and efficient correction for population stratification in multi-locus genome-wide association studies

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Population stratification (PS) is a major confounder in genome-wide association studies (GWAS), leading to false discoveries.
• Existing methods like genomic control (GC), EIGENSTRAT, GMMAT, and LASSOMM have limitations, including arbitrary thresholds or high computational cost.

Purpose of the Study:

• To introduce PCA-LASSO, a novel multi-marker approach for GWAS that addresses the challenges of PS.
• To evaluate the performance of PCA-LASSO against existing single-marker and multi-marker methods.

Main Methods:

• Proposed PCA-LASSO, combining Principal Component Analysis (PCA) for PS correction and Least Absolute Shrinkage and Selection Operator (LASSO) with cross-validation for marker selection.
• Compared PCA-LASSO with single-marker methods (GC, EIGENSTRAT, GMMAT) and a multi-marker method (LASSOMM).

Main Results:

• PCA-LASSO achieved a superior balance between precision and recall (F1 scores) compared to single-marker methods.
• PCA-LASSO significantly improved precision while maintaining recall compared to LASSOMM, with over 1000-fold reduction in computational time.
• Applied to Alzheimer's disease data, PCA-LASSO identified the known risk SNP rs429358 (APOE4).

Conclusions:

• PCA-LASSO offers a simple, fast, and accurate solution for GWAS in the presence of latent population stratification.
• The method provides improved statistical power and efficiency for large-scale genetic association studies.