Testing in semiparametric models with interaction, with applications to gene-environment interactions.

Area of Science:

• Statistics
• Genetics
• Biostatistics

Background:

• Gene-environment interactions are crucial for complex traits.
• Existing statistical tests may be biased in semiparametric models.
• Accurate methods are needed for genetic association studies.

Purpose of the Study:

• Develop unbiased score tests for gene-environment interaction in semiparametric regression.
• Address limitations of existing methods, including bias and computational complexity.
• Provide a computationally feasible and efficient statistical approach.

Main Methods:

• Developed profiled score statistics for semiparametric regression models with Tukey-type interactions.
• Addressed bias and computational challenges of existing score tests.
• Utilized standard bandwidth selection and computational methods for implementation.

Main Results:

• The proposed profiled score statistics are unbiased and asymptotically efficient.
• The new method overcomes computational difficulties associated with integral equations.
• Simulation studies demonstrate improved type I error control and power compared to naive tests.

Conclusions:

• The developed profiled score statistics offer a valid and efficient approach for testing gene-environment interactions.
• This methodology enhances the analysis of complex traits in genetic association studies.
• The approach is illustrated with a case-control study of colorectal adenoma and NAT2 gene.