Predicting the difficulty of pure, strict, epistatic models: metrics for simulated model selection

Area of Science:

• Genetics
• Computational Biology
• Statistical Genetics

Background:

• Simulated datasets are crucial for evaluating genetic disease association algorithms.
• Current evaluations often overlook model architecture's impact on detection difficulty.
• A reliable metric is needed to account for model architecture in simulations.

Purpose of the Study:

• To identify and evaluate metrics for quantifying genetic model detection difficulty.
• To improve the design of simulation studies for genetic association analysis.
• To enhance the selection of genetic models based on architecture.

Main Methods:

• Evaluated Penetrance table variance (PTV), customized odds ratio (COR), and Ease of Detection Measure (EDM) as predictors of detection difficulty.
• Assessed metric reliability across three distinct data search algorithms capable of detecting epistasis.
• Calculated EDM from penetrance values and genotype frequencies.

Main Results:

• EDM and COR were found to be stronger predictors of model detection success than heritability.
• The study identified metrics that quantify model detection difficulty.
• These metrics enable intelligent selection of models from potential architectures.

Conclusions:

• The study formally identifies and evaluates metrics for quantifying genetic model detection difficulty.
• These metrics facilitate improved simulation study design by accounting for architecture-specific detection challenges.
• EDM and COR calculations are now integrated into GAMETES for efficient generation of epistatic models.