Variable selection method for the identification of epistatic models.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Standard genome-wide association studies (GWAS) struggle with complex disease models involving gene interactions.
• Gene interactions are crucial for understanding the heritability of complex human traits.
• Machine learning, like Random Forests (RF), offers potential for detecting interactions but lacks standardized variable selection.

Purpose of the Study:

• To introduce and evaluate a novel variable selection method, relative recurrency variable importance metric (r2VIM).
• To assess r2VIM's performance in identifying epistatic effects (gene-gene interactions) with minimal marginal effects.
• To compare r2VIM against traditional methods like logistic regression in scenarios with large numbers of genetic variants.

Main Methods:

• Development of the relative recurrency variable importance metric (r2VIM).
• r2VIM utilizes recurrency and variance estimation for optimal threshold selection.
• Performance evaluation using simulated data with predominantly epistatic effects.

Main Results:

• r2VIM successfully identifies interaction effects even when marginal effects are virtually absent.
• The method demonstrates robustness with appropriate parameter tuning.
• r2VIM significantly outperforms logistic regression in detecting interactions under these specific conditions.

Conclusions:

• The developed r2VIM method is effective for detecting gene interactions in complex diseases.
• r2VIM provides a powerful alternative to standard GWAS and logistic regression for models with epistatic effects.
• This approach enhances the ability to uncover the genetic architecture of complex human traits.