Association test based on SNP set: logistic kernel machine based test vs. principal component analysis.

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Genome-Wide Association Studies (GWAS) identify risk single nucleotide polymorphisms (SNPs) for complex diseases.
• Individual SNP analysis in GWAS suffers from low power and reproducibility due to multiple comparison corrections.
• SNP set-based methods leverage biological knowledge and genomic features to enhance discovery power.

Purpose of the Study:

• To compare the performance of linear kernel machine (LKM) and principal component analysis (PCA) for SNP set-based association testing.
• To evaluate these methods under various scenarios, including different numbers of causal SNPs and linkage disequilibrium (LD) structures.
• To assess the power of LKM and PCA in detecting multiple causal SNPs and compare them to individual SNP analysis.

Main Methods:

• Utilized simulated datasets with 0 to 3 causal SNPs and simple/complex LD structures.
• Applied linear kernel machine (LKM) tests.
• Employed principal component analysis (PCA) based approaches, varying the number of principal components (PCs).

Main Results:

• Both LKM and PCA maintained type I error control at a 0.05 significance level across simulations.
• PCA with few PCs and LKM are valid for strong LD scenarios.
• Complex LD structures necessitate more PCs for PCA to capture causal SNP information.
• LKM and PCA demonstrated the ability to combine information from multiple causal SNPs, increasing power over single SNP analysis.

Conclusions:

• LKM and PCA are robust methods for SNP set-based association analysis in GWAS.
• The choice of parameters (e.g., number of PCs) for PCA depends on the complexity of the LD structure.
• These methods offer improved power for identifying genetic risk factors in complex diseases compared to traditional single SNP approaches.
• Applied LKM and PCA to a non-small cell lung cancer GWAS dataset, demonstrating real-world applicability.