A penalized robust semiparametric approach for gene-environment interactions

Area of Science:

• Genetics
• Genomics
• Biostatistics

Background:

• Gene-environment (G×E) interactions are crucial in genetic and genomic studies.
• Existing methods for G×E interaction analysis have limitations, including analyzing few gene factors, assuming linear environmental effects, and susceptibility to data contamination.

Purpose of the Study:

• To propose a novel penalized estimation approach for identifying significant G×E interactions.
• To develop a method that jointly models all environmental (E) and gene (G) factors and their interactions, accommodating nonlinear effects and data contamination.

Main Methods:

• A partially linear varying coefficient model is employed to handle nonlinear effects of environmental factors.
• A rank-based loss function is utilized to address potential data contamination.
• Penalization techniques are applied for variable selection in high-dimensional data, enabling automatic determination of G factor effects (interaction, main effect, or no effect).
• A coordinate descent algorithm is used for efficient implementation.

Main Results:

• The proposed penalized estimation approach demonstrates satisfactory performance in simulations.
• The method outperforms several competing alternative approaches in identifying G×E interactions.
• The approach was successfully applied to analyze a lung cancer study using gene expression data and clinical variables.

Conclusions:

• The developed penalized estimation approach offers an effective and robust method for identifying G×E interactions.
• This approach overcomes limitations of existing methods by jointly modeling factors, accommodating nonlinearity, and handling data contamination.
• The method has practical utility, as demonstrated by its application in a real-world lung cancer study.