RL-SKAT: An Exact and Efficient Score Test for Heritability and Set Tests

Area of Science:

• Statistical Genetics
• Bioinformatics
• Genomic Association Studies

Background:

• Testing variance components in linear mixed models is crucial for statistical genetics.
• The score test is vital for genetic marker-phenotype association and heritability testing.
• The Sequence Kernel Association Test (SKAT) is a popular score-based method, but its P-value calibration can be problematic in small samples or with unrelated individuals.

Purpose of the Study:

• To characterize the conditions leading to P-value discrepancies in SKAT.
• To develop an efficient method for calculating exact P-values for the score test.
• To improve the accuracy and power of heritability and set-based association tests.

Main Methods:

• Analysis of score test statistic distribution under various conditions.
• Development of an efficient algorithm for exact P-value computation for single variance component models with continuous responses.
• Validation using simulated and real-world large-scale genetic datasets (e.g., Wellcome Trust Case Control Consortium 2).

Main Results:

• Identified conditions where SKAT's P-value approximation is inaccurate, leading to overconservative and underpowered tests, particularly with unrelated individuals.
• Demonstrated that the proposed method significantly speeds up P-value calculation by orders of magnitude.
• Showcased the accurate P-value estimation provided by the new method in both simulated and real datasets.

Conclusions:

• The developed method provides fast and accurate P-values for score tests, addressing SKAT's calibration issues.
• This advancement enables more reliable and powerful genetic association studies, including heritability and set-based analyses.
• The method is publicly available, facilitating broader application in genomic research.