Linear score tests for variance components in linear mixed models and applications to genetic association studies

Area of Science:

• Genetics
• Statistical genomics
• Pharmacogenomics

Background:

• Genome-scale genotyping technologies enable genetic association mapping for identifying disease-related genomic regions.
• Early-phase pharmacogenomic studies often have limited sample sizes, requiring accurate and efficient statistical tests.
• Existing methods may not fully address the need for tests applicable to complex genetic models, including interactions and non-linearity.

Purpose of the Study:

• To develop a novel score-based genetic association test for detecting genomic regions associated with phenotypes.
• To ensure the test is accurate in small samples, computationally fast, and powerful for complex genetic models.
• To provide a viable solution within the kernel machine framework for genetic association studies.

Main Methods:

• Utilized kernel machine methods within linear mixed models, transforming the problem into testing variance component nullity.
• Developed score-based tests using a statistic linear in the score function.
• Introduced a new moment-based approximation for null models with multiple variance parameters.

Main Results:

• The proposed score-based test is exact in finite samples when the null model has a single error variance parameter.
• The moment-based approximation performs well in simulations for more complex null models.
• The new test demonstrates accuracy, computational efficiency, and power comparable to or exceeding existing methods.

Conclusions:

• The developed score-based test offers a robust and versatile tool for genetic association mapping, particularly in pharmacogenomics.
• The method addresses key requirements for modern genetic studies, including handling complex genetic architectures and small sample sizes.
• This approach provides a valuable alternative to existing quadratic score tests and restricted likelihood ratio tests.