Genome-wide association analysis by lasso penalized logistic regression

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Ordinary regression with lasso penalty simplifies continuous model selection.
• Lasso penalized regression is effective when predictors outnumber observations.
• This study focuses on disease gene mapping using genetic markers.

Purpose of the Study:

• Evaluate lasso penalized logistic regression for case-control disease gene mapping.
• Assess the performance with a large number of single nucleotide polymorphisms (SNPs) as predictors.
• Investigate the identification of relevant SNPs and their interactions.

Main Methods:

• Utilized lasso penalized logistic regression for SNP selection.
• Tuned lasso penalty strength to predetermine the number of relevant SNPs.
• Employed cyclic coordinate ascent for efficient penalized likelihood maximization.
• Examined two-way and higher-order interactions among selected SNPs.

Main Results:

• Demonstrated the effectiveness of the strategy on simulated and real data.
• Replicated previous SNP findings for coeliac disease.
• Provided insights into potential SNP interactions relevant to coeliac disease.

Conclusions:

• Lasso penalized logistic regression is a powerful tool for disease gene mapping with high-dimensional SNP data.
• The method facilitates the identification of key genetic markers and their complex interactions.
• This approach aids in understanding the genetic architecture of diseases like coeliac disease.