Evaluation of Existing Methods for High-Order Epistasis Detection.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Understanding complex traits requires identifying epistatic interactions among genetic loci.
• Numerous epistasis detection methods exist, complicating selection for Genome-Wide Association Studies (GWAS).
• Evaluating these methods is essential for advancing genetic architecture studies.

Purpose of the Study:

• To compare epistasis detection methods based on runtime, detection power, and type I error rate.
• To specifically assess performance for high-order genetic interactions.
• To guide researchers in selecting appropriate methods for their GWAS.

Main Methods:

• Comparative analysis of various epistasis detection algorithms.
• Evaluation across different experimental conditions, including presence/absence of marginal effects.
• Assessment of computational cost (runtime) and statistical performance (detection power, false positives).

Main Results:

• Exhaustive methods demonstrated superior detection power across all tested scenarios.
• Non-exhaustive methods showed inconsistent performance, particularly when marginal genetic effects were absent.
• Methods like BADTrees, FDHE-IW, SingleMI, and SNPHarvester performed well for high-order interactions when marginal effects were present.
• SNPHarvester, FDHE-IW, and DCHE exhibited strong control over false positives.

Conclusions:

• No single epistasis detection method is universally optimal for all GWAS scenarios.
• Exhaustive methods are recommended when computational resources permit, especially for large datasets.
• Non-exhaustive methods are viable alternatives when computational time is a limiting factor.
• Method selection should balance detection power, error rates, and resource availability.