Robust kernel association testing (RobKAT).

Area of Science:

• Genetics
• Statistical genetics
• Bioinformatics

Background:

• Kernel machine methods, such as the Sequence Kernel Association Test (SKAT), are commonly used for testing single-nucleotide polymorphism (SNP) associations with responses.
• Existing methods often rely on least-squares procedures, assuming normally distributed responses, which limits their applicability.
• Robust methods like Quantile Regression Kernel Machine (QRKM) have limitations in loss function flexibility and inference scope.

Purpose of the Study:

• To propose a general and robust kernel association test (RobKAT) that overcomes the limitations of existing methods.
• To develop a test with a flexible choice of loss function and no distributional assumptions on the response.
• To evaluate the performance of RobKAT through simulations and real-world data application.

Main Methods:

• Developed RobKAT, a novel robust kernel association test.
• Evaluated RobKAT's type I error control and power through simulations across various data distributions.
• Applied RobKAT to analyze gene-environment associations in schizophrenia patients using data from the Clinical Antipsychotic Trials of Intervention Effectiveness (CATIE) study.

Main Results:

• RobKAT controls type I error and shows comparable power to SKAT under normal error distributions.
• In non-normal error distributions, RobKAT demonstrated similar or greater power than SKAT.
• RobKAT identified significant associations with four SNP sets in the CATIE data, including three missed by SKAT.

Conclusions:

• RobKAT provides a flexible and robust approach for genetic association testing, suitable for diverse data distributions.
• The proposed method enhances the ability to detect genetic associations, particularly in complex datasets.
• RobKAT offers a valuable alternative to existing methods, improving power and applicability in genetic association studies.