Dissecting gene-environment interactions: A penalized robust approach accounting for hierarchical structures

Area of Science:

• Genomics and Bioinformatics
• Cancer Research
• Statistical Genetics

Background:

• Identifying gene-environment (G x E) interactions is crucial for understanding complex diseases like cancer, but presents significant challenges in high-throughput studies.
• Existing methods often fail to account for the joint effects of numerous genetic variants, hierarchical structures of main effects and interactions, data contamination, or employ inefficient selection techniques.
• These limitations hinder accurate identification of G x E interactions and main effects in large-scale genetic datasets.

Purpose of the Study:

• To develop an effective penalization approach for identifying significant gene-environment (G x E) interactions and main genetic effects.
• To create a method that respects the hierarchical structure of main effects and interactions.
• To accommodate potential data contamination and utilize efficient selection techniques for complex sparse data.

Main Methods:

• Development of a novel penalization approach designed to identify important G x E interactions and main effects.
• Incorporation of the least absolute deviation (LAD) loss function to robustly handle potential data contamination.
• Application of the method under the accelerated failure time (AFT) model for survival data analysis.

Main Results:

• The proposed penalization approach effectively identifies important G x E interactions and main effects, respecting hierarchical structures.
• The method demonstrates robustness against data contamination due to the use of the LAD loss function.
• Simulations and a lung cancer prognosis case study confirm the superiority of the proposed approach over existing alternatives.

Conclusions:

• The developed penalization method offers an effective solution for identifying gene-environment interactions and main effects in high-throughput cancer studies.
• This approach improves upon existing methods by addressing hierarchical structures, data contamination, and complex sparsity.
• The findings have significant implications for understanding the genetic and environmental components of cancer prognosis.