Interaction models matter: an efficient, flexible computational framework for model-specific investigation of...

Area of Science:

• Genetics and Genomics
• Computational Biology
• Systems Biology

Background:

• Epistasis, the interaction between multiple genes, is fundamental to genetics but often understudied due to computational demands and a focus on single-locus effects.
• Current epistasis detection methods predominantly use Cartesian (multiplicative) models, potentially overlooking complex, non-linearly separable genetic relationships.
• The limited scope of existing models may hinder the comprehensive exploration of epistatic interactions that drive biological phenomena.

Purpose of the Study:

• To develop novel algorithms for detecting epistasis that accommodate diverse interaction models beyond the standard Cartesian approach.
• To enable efficient computational analysis of two-way, three-way, and higher-order epistasis with improved memory usage.
• To investigate the impact of using varied epistasis models, including the XOR model, on identifying biologically relevant genetic interactions in mammalian datasets.

Main Methods:

• Introduction of new algorithms for calculating interaction coefficients in regression models, supporting multiple interaction term models for genetic loci.
• Development of statistical tests for interaction coefficients and an efficient matrix-based algorithm for permutation testing of two-way epistasis.
• Application of the developed algorithms to rat and mouse datasets (10,000+ loci, 1,000+ samples) using both Cartesian and XOR models to analyze body mass index.

Main Results:

• Significant epistatic loci overlap between models in rats, but the specific interacting pairs were largely distinct, highlighting model-dependent discoveries.
• The XOR model demonstrated substantially greater evidence for statistical epistasis across numerous locus pairs in both rat and mouse datasets.
• In rats, loci identified through XOR-based epistasis analysis were enriched for pathways with known biological relevance.

Conclusions:

• The study underscores that relying on a single epistasis model can lead to the omission of numerous biologically significant genetic relationships.
• Implementing diverse interaction models is essential for a comprehensive understanding of the complex epistatic interactions present in living organisms.
• The findings advocate for a broader methodological approach in genetic research to fully capture the landscape of gene-gene interactions.