Alternative methods for H1 simulations in genome-wide association studies

Area of Science:

• Genetics
• Statistical genetics
• Computational biology

Background:

• Statistical power is crucial for genome-wide association (GWA) studies.
• Empirical power estimation involves simulating phenotypes under a disease model (H1).
• Current gold standard methods (e.g., Hapgen) simulate genotypes based on phenotypes.

Purpose of the Study:

• Introduce a novel, faster approach for simulating phenotypes under H1.
• Avoids the need for generating new genotypes for each simulation.
• Provides a flexible and efficient alternative for power assessment in GWA studies.

Main Methods:

• Developed three algorithms: rejection sampling, Markov chain Monte Carlo (MCMC), and backward sampling.
• Validated algorithms on simulated and realistic datasets, comparing with Hapgen.
• Applied the method to a 1000 Genomes Project dataset (629 individuals, 8,048 SNPs on chromosome X) with an additive model and epistasis.

Main Results:

• All three algorithms yielded consistent results, with backward sampling being significantly faster.
• The proposed method produced results comparable to Hapgen.
• Epistatic effects were shown to be significant even with simple marker statistics.
• GWA study performance is highly dependent on disease prevalence, with higher prevalence improving power.

Conclusions:

• The developed approach is a viable and faster alternative to Hapgen-type methods.
• Advantages include no need for complex genotype models (haplotypes, recombination rates).
• Offers unconstrained disease model selection (SNPs, gene-environment interactions, hybrid models).
• Algorithms are available in the R package 'waffect'.