Integrative analysis of gene-environment interactions under a multi-response partially linear varying coefficient...

Area of Science:

• Genetics
• Biostatistics
• Bioinformatics

Background:

• Analyzing genetic data with multiple correlated responses presents challenges in identifying gene-environment (GxE) interactions and main effects.
• Existing models may not fully capture complex genetic architectures, including nonlinear environmental effects or both homogeneous and heterogeneous genetic bases for multiple traits.

Purpose of the Study:

• To develop an integrative statistical method for analyzing genetic data with multiple correlated response variables.
• To simultaneously identify important gene-environment (GxE) interactions, main gene effects, and main environment effects.
• To accommodate potential nonlinear environmental effects and both homogeneous and heterogeneous genetic models.

Main Methods:

• A multi-response partially linear varying coefficient model was employed to handle complex relationships and nonlinear environmental effects.
• Penalization techniques were utilized for efficient marker selection, capable of identifying variants with or without GxE interactions and main effects.
• A coordinate descent algorithm was implemented for effective computation of the proposed model.

Main Results:

• The proposed penalization method demonstrated the ability to select genetic variants involved in GxE interactions, main gene effects, main environment effects, or combinations thereof.
• The method successfully accommodated both homogeneity and heterogeneity models for the genetic basis of multiple responses.
• Simulation studies and analysis of the Health Professionals Follow-up Study data showed superior performance compared to existing methods.

Conclusions:

• The developed penalization method provides a powerful tool for integrative analysis of genetic data with multiple correlated responses.
• It effectively identifies complex GxE interactions and main effects, offering improved insights into genetic architectures.
• The method's robustness and superior performance are validated through simulations and real-world data application.