FastSKAT: Sequence kernel association tests for very large sets of markers

Area of Science:

• Genetics and Genomics
• Statistical Genetics
• Bioinformatics

Background:

• The sequence kernel association test (SKAT) is a standard method for analyzing associations between phenotypes and rare genetic variants.
• SKAT's computational performance is limited by the eigenvalue decomposition of an n x n genotype covariance matrix, especially for large sample sizes (n > 10^4).
• This computational bottleneck hinders the practical application of SKAT in large-scale genetic studies.

Purpose of the Study:

• To develop a computationally efficient approximation for SKAT's null distribution tail probabilities.
• To overcome the computational limitations of traditional SKAT for large sample sizes.
• To enable new applications of SKAT by reducing computational burden.

Main Methods:

• Proposed fastSKAT, an approximation method that extracts the k largest eigenvalues of a weighted genotype covariance matrix.
• Utilizes the Satterthwaite approximation for the remaining eigenvalues, significantly reducing computational complexity.
• Provides guidance on selecting the parameter k and identifies scenarios where its choice impacts results.

Main Results:

• fastSKAT provides accurate approximations to tail probabilities with substantially reduced computational cost compared to standard SKAT.
• The method demonstrates improved efficiency, making SKAT feasible for larger datasets.
• The approach facilitates novel applications, such as grouping variants by topological domains or comparing chromosome-wide associations based on histone marker classes.

Conclusions:

• fastSKAT offers a computationally inexpensive and accurate alternative for evaluating SKAT's null distribution.
• This advancement significantly enhances the scalability and applicability of SKAT in genetic association studies.
• The method opens new avenues for exploring complex genetic architectures and genotype-phenotype relationships.