Tradeoffs in Modeling Context Dependency in Complex Trait Genetics.

Area of Science:

• Genetics
• Statistical Genomics
• Complex Trait Analysis

Background:

• Genetic effects on complex traits can be influenced by environmental context (Gene-by-Environment interaction, GxE).
• The utility of complex GxE models versus simpler additive models in genome-wide association studies (GWAS) remains debated.
• Standard GWAS typically employ additive models, potentially overlooking crucial context-dependent genetic effects.

Purpose of the Study:

• To propose a decision framework for selecting between additive and GxE models in GWAS based on bias-variance tradeoffs.
• To evaluate the utility of considering context dependency in genetic analyses.
• To demonstrate how polygenic GxE models can enhance trait estimation and prediction.

Main Methods:

• Derived a decision rule balancing estimation noise against bias reduction for model selection.
• Analyzed Gene-by-Sex (GxSex) interactions in human physiology data.
• Examined Gene-by-Diet (GxDiet) effects on fruit fly longevity, focusing on polygenic patterns.

Main Results:

• Independent GxE analyses (e.g., GxSex) often show increased estimation noise that outweighs bias reduction.
• Polygenic GxE models, considering context dependency across multiple variants, effectively mitigate both noise and bias.
• Polygenic GxE analysis improves both the estimation of genetic effects and prediction accuracy for complex traits.

Conclusions:

• GxE models are not always superior to additive models when variants are analyzed independently.
• Joint consideration of polygenic GxE patterns is crucial for accurate interpretation and improved prediction of complex traits.
• Focusing solely on 'top hits' from independent analyses can be misleading; polygenic GxE provides a more robust approach.