Dimension Reduction Using Local Principal Components for Regression-Based Multi-SNP Analysis in 1000 Genomes and the...

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Multi-collinearity poses a significant challenge in genetic association studies using multiple Single Nucleotide Polymorphisms (SNPs).
• This issue can lead to regression model instability and failure in genetic association analyses.
• Existing methods struggle to adequately address severe multi-collinearity in dense genotype data.

Purpose of the Study:

• To propose and evaluate a novel dimension reduction method, Dimension Reduction using Local Principal Components (DRLPC), for addressing multi-collinearity.
• To improve the power and applicability of regression-based statistical tests in genetic association studies.
• To assess the effectiveness of DRLPC in reducing variable numbers while preserving essential genetic information.

Main Methods:

• DRLPC removes SNPs exhibiting high linear dependency, assuming remaining SNPs capture their effects.
• Variance Inflation Factor (VIF) is used to quantify collinearity, with SNPs above a VIF threshold (e.g., 20) being excluded.
• The method was applied to chromosome 22 SNPs from the 1000 Genomes Project and the Canadian Longitudinal Study on Aging (CLSA).

Main Results:

• DRLPC significantly reduces the number of SNPs for regression analysis, particularly for larger genes (average reduction to ~20%).
• For smaller genes, the reduction is less pronounced (average ~48%), indicating tailored effectiveness.
• Application of DRLPC improved the power of the multiple regression Wald test from 60% to approximately 80% in simulation studies.

Conclusions:

• DRLPC is an effective strategy for mitigating multi-collinearity in genetic association studies.
• The method enhances the power of statistical tests, leading to more robust genetic association findings.
• DRLPC offers improved applicability for subsequent regression analyses, especially with large genetic datasets.