Practical issues in screening and variable selection in genome-wide association analysis.

Area of Science:

• Genetics
• Statistical Modeling
• Bioinformatics

Background:

• High-dimensional data analysis is crucial for statistical modeling, especially in genome-wide association studies (GWAS).
• Computational cost and estimation accuracy are key challenges in analyzing ultrahigh-dimensional data like single nucleotide polymorphisms (SNPs).
• Current GWAS methods often use a two-stage approach: pre-screening followed by penalized regression for variable selection.

Purpose of the Study:

• To investigate the performance of different pre-screening methods combined with penalized regression techniques.
• To determine the optimal combination for identifying causal SNPs in quantitative trait genome-wide association studies (GWAS).
• To evaluate these methods using two real-world GWAS datasets.

Main Methods:

• Utilized a two-stage variable selection approach.
• Employed various pre-screening techniques based on P-values and regression coefficients.
• Applied penalized regression models including ridge, lasso, adaptive lasso, and elastic-net for variable selection.

Main Results:

• Compared the efficacy of different pre-screening and penalized regression combinations.
• Assessed performance on quantitative phenotypes using two distinct genome-wide association study (GWAS) datasets.
• Identified specific combinations that offer superior performance in terms of accuracy and efficiency.

Conclusions:

• The choice of pre-screening method significantly impacts the performance of penalized regression in GWAS.
• Certain combinations of pre-screening and penalized regression are more effective for variable selection in ultrahigh-dimensional genomic data.
• Findings provide guidance for optimizing statistical analysis in genome-wide association studies (GWAS).