Optimal designs for estimating and testing interaction among multiple loci in complex traits by a Gibbs sampler

Area of Science:

• Genetics
• Statistical Genetics
• Bioinformatics

Background:

• Estimating genetic interactions is crucial for understanding complex traits.
• The Bayesian approach using Gibbs sampling (BAGS) is a novel method for such estimations.
• Practical guidelines for experimental design in BAGS are needed.

Purpose of the Study:

• To provide practical guidelines for experimental designs using the Bayesian approach using Gibbs sampling (BAGS).
• To determine optimal sample sizes and design parameters for estimating genetic interactions.
• To evaluate the performance of BAGS across various simulation scenarios.

Main Methods:

• Simulated diverse datasets varying in number of loci, within-genotype variance, sample size, and design balance.
• Employed the Bayesian approach using Gibbs sampling (BAGS) for interaction estimation.
• Calculated Mean Square Prediction Error (MSPE) and empirical statistical power.

Main Results:

• Optimal sample sizes were determined for 2-, 3-, and 4-locus unbalanced data to achieve MSPE > 2.0 and power > 0.8.
• A strong negative correlation (-0.8) between MSPE and power suggests simultaneous consideration for optimal design.
• Specific sample sizes (135, 675, >8100) were identified for different locus numbers under specific variance conditions.

Conclusions:

• The Bayesian approach using Gibbs sampling (BAGS) is suitable for detecting interaction effects among a limited number of loci (≤4).
• Simultaneous optimization of MSPE and statistical power is recommended for robust experimental design.
• A practical guideline is established for determining optimal sample sizes based on desired power levels or vice versa for BAGS.