GenEpi

GenEpi identifies epistatic interactions between genetic variants associated with phenotypes using machine learning, a two-stage modeling workflow, two-element combinatorial encoding, and L1-regularized regression with stability selection.

Key Features:

• Two-Stage Modeling Workflow: Uncovers both within-gene and cross-gene epistasis through a two-stage modeling approach.
• Two-Element Combinatorial Encoding: Uses two-element combinatorial encoding to produce features that capture potential interactions between genetic variants.
• L1-Regularized Regression with Stability Selection: Builds predictive models using L1-regularized regression combined with stability selection to identify significant variant interactions while reducing overfitting.
• Machine Learning Integration: Integrates machine learning techniques with encoding and regression to model complex epistatic effects.
• GWAS and Dataset Applicability: Addresses limitations of Genome-Wide Association Studies (GWAS) in detecting variant interactions and applies to both simulated and real-world datasets.

Scientific Applications:

• Alzheimer's disease (AD): Identifies disease-related variants and epistatic interactions with biological relevance and predictive power in AD datasets.
• Complex multifactorial disease genetics: Enables investigation of genetic interactions underlying multifactorial conditions to inform studies of pathogenesis.

Methodology:

Applies a two-stage modeling workflow with two-element combinatorial encoding to generate features, then fits L1-regularized regression models with stability selection within a machine learning framework to detect epistasis in simulated and real-world datasets.