Screen and clean: a tool for identifying interactions in genome-wide association studies

Area of Science:

• Genetics
• Statistical genomics
• Computational biology

Background:

• Epistasis, or gene-gene interactions, is a potential source of disease risk.
• Discovering interacting loci in genome-wide association studies (GWAS) remains challenging.
• Current GWAS methods often focus on marginal tests for individual pairwise interactions.

Purpose of the Study:

• To explore the Screen and Clean (SC) procedure for identifying disease liability loci, including interactions.
• To utilize the lasso procedure, a model selection tool for high-dimensional regression, within the SC framework.
• To develop and evaluate algorithms for implementing SC in GWAS.

Main Methods:

• The SC procedure uses a varying dictionary approach with the lasso for model selection.
• Initial screening identifies promising single-nucleotide polymorphisms (SNPs) based on main effects.
• The dictionary is updated to include interaction terms, followed by further lasso screening and a cleaning process to identify significant terms.

Main Results:

• The SC method successfully controls Type I error rates.
• SC demonstrates good power in identifying risk loci and their interactions.
• Application to Type 1 Diabetes data revealed evidence of interactions within the HLA class II region and Chromosome 12q24.

Conclusions:

• The SC procedure offers an effective approach for discovering genetic interactions in GWAS.
• SC provides a robust method for analyzing complex genetic architectures underlying diseases.
• The findings highlight potential interacting regions associated with Type 1 Diabetes risk.