Conditional asymptotic inference for the kernel association test.

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• The kernel association test (KAT) is widely used in biological studies to detect genetic associations, especially for combining weak or opposing effects.
• Current P-value assessment for KAT relies on asymptotic distributions, which are limited to specific trait types (continuous, dichotomous) and can be conservative with small sample sizes.
• Permutation tests offer an alternative for exact inference but are computationally intensive, posing practical challenges in large-scale genetic analyses.

Purpose of the Study:

• To introduce a novel conditional asymptotic distribution for the kernel association test (KAT) applicable to all trait types.
• To provide an efficient computational method for P-value calculation, addressing the limitations of existing approaches.
• To validate the new method's performance through simulations and real genetic data analysis.

Main Methods:

• Derived a conditional asymptotic distribution for the KAT based on prior theoretical work.
• Developed an explicit formula for computing P-values from this distribution.
• Validated the method using extensive simulations with real genotype data and analyzed data from the Ocular Hypertension Treatment Study.

Main Results:

• The new conditional asymptotic distribution provides a valid approximation for KAT P-values across all trait types.
• The method effectively controls the type I error rate and demonstrates slight conservativeness compared to permutation tests.
• Simulation and real data analyses confirm the approach's utility and computational efficiency.

Conclusions:

• The proposed conditional asymptotic distribution offers a computationally efficient alternative for assessing KAT P-values.
• This method can serve as a fast screening tool, enabling targeted use of more time-consuming permutation tests for significant findings.
• The R package iGasso provides an implementation of this novel approach for broader accessibility in genetic research.