FastSKAT

FastSKAT accelerates set-based rare-variant association testing by approximating the Sequence Kernel Association Test (SKAT) null distribution through extraction of leading eigenvalues (or singular values) from a weighted genotype covariance matrix and estimating remaining eigenvalues with a single-term Satterthwaite approximation.

Key Features:

• SKAT association testing: Implements the Sequence Kernel Association Test (SKAT) to assess associations between phenotypes and sets of genetic variants, with emphasis on rare variants.
• Eigenvalue-based computation: Operates on eigenvalues of a genotype covariance matrix, noting the conventional cubic computational complexity with respect to sample size.
• Leading-eigenvalue extraction: Extracts only the largest eigenvalues (or singular values) from a weighted genotype covariance matrix to reduce computational demands.
• Satterthwaite approximation: Estimates the remaining eigenvalues using a single-term Satterthwaite approximation to approximate tail probabilities or quantiles of the null distribution.
• Parameter k flexibility: Supports a user-specified parameter k (number of largest eigenvalues extracted) and includes guidance and alerts when the choice may influence results.
• Scalability for large cohorts: Addresses computational bottlenecks that arise for sample sizes exceeding 10,000 by reducing the cost of full eigenvalue decomposition.
• Extended analysis applications: Enables analyses such as grouping variants by topologically assisted domains and chromosome-wide association studies based on classes of histone markers.

Scientific Applications:

• Rare variant association testing: Performs set-based tests for associations involving rare genetic variants.
• Large-scale genomic studies: Facilitates SKAT analyses in cohorts with sample sizes greater than 10,000 where full eigen decomposition is computationally prohibitive.
• Topologically assisted domain analyses: Supports grouping of genetic variants by topologically assisted domains for association testing.
• Histone marker–based studies: Enables chromosome-wide association studies stratified by classes of histone markers.
• Complex phenotype association: Applied to studies of complex phenotypic associations involving sets of variants.

Methodology:

Computes eigenvalues of a genotype covariance matrix, extracts the largest eigenvalues (or singular values) from a weighted genotype covariance matrix, approximates tail probabilities or quantiles of the null distribution by estimating remaining eigenvalues with a single-term Satterthwaite approximation, and thereby reduces the need for full cubic-complexity eigen decomposition.