Two-stage strategies to detect gene x gene interactions in case-control data

Area of Science:

• Genetics
• Computational Biology
• Statistical Genetics

Background:

• Large-scale genetic association studies are vital for complex disease research.
• Evaluating gene x gene interactions (epistasis) is key to understanding genetic effects.
• Computational burden limits comprehensive pairwise SNP interaction testing.

Purpose of the Study:

• To compare two-stage strategies for pairwise SNP interaction testing.
• To identify efficient methods for reducing computational load in epistasis analysis.
• To assess the impact of SNP selection criteria on interaction detection.

Main Methods:

• Comparison of simultaneous and conditional two-stage SNP interaction testing approaches.
• Stage 1: SNP selection based on marginal significance thresholds (p=0.05, p=0.1) or Bonferroni-adjusted significance.
• Stage 2: Pairwise interaction testing on selected SNPs using simulated data (GAW15 Problem 3).

Main Results:

• Most detected interactions involved SNP pairs within 1000 kb.
• False positives arose from SNPs with strong marginal signals.
• Locus proximity and marginal signal strength are critical factors in interaction evaluation.

Conclusions:

• Two-stage strategies effectively reduce computational burden in SNP interaction testing.
• Accounting for locus proximity is essential for accurate epistasis analysis.
• Marginal signal strength is important for logistic regression-based interaction modeling; additive effects can capture dominance interactions.