EMLRT

EMLRT performs association testing that integrates genotype imputation uncertainty to improve statistical power and control type I error for markers across a range of minor allele frequencies (MAF).

Key Features:

• Incorporation of Imputation Uncertainty: Integrates genotype imputation uncertainty into association testing using posterior genotype probabilities or imputed dosages.
• Methodological Approach: Supports two scenarios: Scenario I with posterior probabilities for all potential genotypes and Scenario II with only imputed dosage summary statistics.
• Expectation-Maximization Likelihood-Ratio Test (EM-LRT): Implements an EM-LRT that operates on posterior genotype probabilities in Scenario I and on probabilities derived from dosages in Scenario II.
• Genotype Probability Sampling: When only dosages are available, samples genotype probabilities from the posterior distribution conditional on imputed dosage before testing.
• Statistical Robustness: Simulations demonstrate controlled type I error and preserved or improved statistical power across a broad spectrum of MAFs and imputation qualities.
• Comparative Performance: EM-LRT-Prob (Scenario I) provides superior power to traditional methods, and EM-LRT-Dose (Scenario II) matches EM-LRT-Prob and outperforms standard dosage-based methods, particularly for low MAF or poor imputation quality.

Scientific Applications:

• Association testing with imputed genotypes: Applied to association analyses that require incorporation of imputation uncertainty.
• Low MAF marker analysis: Suited for analysis of markers with low minor allele frequency (MAF) where imputation quality is suboptimal.
• Large-scale and cohort studies: Applicable to large-scale genomic studies, diverse populations, and complex trait mapping, with validation on the Cebu Longitudinal Health and Nutrition Survey and the Women's Health Initiative Study.

Methodology:

Implements an expectation-maximization likelihood-ratio test (EM-LRT); uses posterior genotype probabilities in Scenario I and samples genotype probabilities from the posterior conditional on imputed dosage in Scenario II before applying EM-LRT; performance was evaluated via simulations assessing type I error and power.