A framework for modeling epistatic interaction.

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Numerous tools exist for detecting epistasis-associated single nucleotide polymorphisms (SNPs).
• However, statistical epistasis models used by these tools have not been independently evaluated.
• This limits understanding of tool performance and hinders development of improved methods.

Purpose of the Study:

• To develop a protocol for evaluating epistasis models independently of detection tools.
• To generalize existing models for dichotomous phenotypes to categorical and quantitative traits.
• To propose a novel maximum likelihood (ML) model for epistasis detection.

Main Methods:

• Developed a protocol for independent evaluation of epistasis models.
• Generalized existing statistical models to handle categorical and quantitative phenotypes.
• Proposed a new ML model scoring SNP sets based on phenotype penetrance distributions.

Main Results:

• The proposed ML model demonstrated superior performance compared to three widely used epistasis models in extensive experiments.
• Quadratic regression models showed strong performance for quantitative phenotypes.
• The study provides insights into the behavior of existing epistasis detection models.

Conclusions:

• The developed protocol enables objective assessment of epistasis models.
• The novel ML model offers improved accuracy for detecting epistasis.
• This work facilitates the development of more effective tools for genetic association studies.